Dataset Viewer
problem_id
stringlengths 8
8
| problem
stringlengths 1.78k
2.3k
| data_type
stringclasses 1
value | problem_type
stringclasses 1
value | options
listlengths 0
0
| solution
stringclasses 102
values | path
stringlengths 25
25
| data_source
stringclasses 1
value |
|---|---|---|---|---|---|---|---|
001510_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A playful kitten pounces on a dangling string, swats it with its paws, then leaps and chases it across the room..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001510_u.mp4
| |
000867_f
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An ultra-relaxing landscape with a lake of crystal-clear water in motion, while splashes of rain make the environment even more magical. Brightly colored flowers dance with the wind..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000867_f.mp4
| |
001047_h
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Imagine an idyllic countryside estate, where rows of Juliet roses bloom in harmony with the landscape. Describe the rolling hills, the gentle rustling of leaves, and the sight of these romantic roses swaying in the gentle breeze..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001047_h.mp4
| |
000892_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A person reading an eBook on a tablet while sitting on a bench in a park with trees and flowers..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000892_m.mp4
| |
004842_p
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A close-up shot of a person repairing a bicycle chain using a chain tool. The person's hands focus on aligning or adjusting the chain on a light green bicycle with a black chain and gears. The scene takes place in a workshop with a blue workbench and various tools in the blurred background. Bright, even lighting highlights the intricate details of the bicycle and the careful movements of the hands. The camera captures the precise action with a stationary, slightly angled view..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 5; physical/common-sense consistency: 2</answer>
|
./vs2_videos/004842_p.mp4
| |
002724_g
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An intricately detailed owl, with feathers that seem to glisten and shimmer in the light, swoops into the scene carrying a letter in its beak. The letter is sealed with a red wax stamp that bears the emblem of Hogwarts..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/002724_g.mp4
| |
001002_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man with a beard and short hair is sitting in the back seat of a car, wearing a black shirt with a graphic design. He starts speaking, then picks up a red megaphone, holding it close to his mouth as he continues to speak into it. His facial expressions and hand gestures show he is explaining something. The camera captures a medium close-up of him, focusing on his upper body and face, with the car interior visible in the background..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001002_m.mp4
| |
002653_c
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A Ferrari Formula 1 car, in white livery, racing around the streets of Monaco, in second place to a Williams car..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/002653_c.mp4
| |
000161_a
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Lights gently blink as the man finishes tuning his guitar. The crowd behind him is gently moving in expectation. He tips his hat to us just before the video is over..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 4; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000161_a.mp4
| |
000933_x
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Two people who meet for the first time and greet each other, one wearing a suit marked Evapco and the other wearing a suit marked with the flag of Japan. A third person stands between them, making the presentation, wearing a suit marked RSI..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000933_x.mp4
| |
000110_a
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man shovels snow off the driveway, clears a path to the door, and then stands back, watching the snow gently fall once more..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000110_a.mp4
| |
000958_x
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man grabs a towel, dries his face and arms, then hangs it over a chair to dry..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 2; physical/common-sense consistency: 1</answer>
|
./vs2_videos/000958_x.mp4
| |
002654_g
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A young man with blonde hair and glasses sits in front of a shelf filled with colorful Disney figurines and collectibles. Wearing a blue Disneyland t-shirt, he appears to be speaking or reacting to something off-camera. The lively Disney-themed background adds a playful atmosphere, while his minimal movements and facial expressions suggest an engaged conversation. The camera remains stationary, focusing on him and the vibrant figurines behind him..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 4; physical/common-sense consistency: 4</answer>
|
./vs2_videos/002654_g.mp4
| |
000890_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A lively classroom scene with a teacher and three students engaged in a playful interaction. The teacher, wearing a white shirt and black vest, holds a globe and gestures animatedly as he speaks. Two girls and a boy, dressed in school uniforms, are seated at desks, smiling and reacting to the teacher's antics. The atmosphere is light-hearted and filled with laughter as they share a fun moment. The camera captures the scene with all subjects visible in a medium shot, highlighting the dynamic conversation and camaraderie..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 1</answer>
|
./vs2_videos/000890_q.mp4
| |
001404_h
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Experience the majesty of bears in their natural environments, showcasing the balance between these creatures and the landscapes they call home..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001404_h.mp4
| |
004974_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
The storm brews as the lighthouse keeper diligently tends to the light, ensuring it remains strong and steady despite the raging storm outside. Crane up..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004974_z.mp4
| |
002677_g
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A sleek drone hovers silently above a dimly lit room, its camera capturing the intricate details of classified documents sprawled across a table in cyborg style..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/002677_g.mp4
| |
001591_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
The scene opens with a wide shot of an ancient palace. The palace is grand and magnificent, with palace maids moving quickly through the hallways, attending to their duties. Tilt down..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001591_z.mp4
| |
001039_k
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A writer is typing on a typewriter, with a sheet of paper showing 'Novel in Progress: 'The Mystery of the Red Mansion', Chapter 7'..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001039_k.mp4
| |
001529_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman with long, wavy blonde hair stands in a cozy room, wearing a sleeveless top. As she speaks or reacts to something, her facial expressions shift from neutral to slightly concerned. The background features a serene setting with a waterfall picture, string lights, and a mirror. Her subtle head movements and occasional glances to the side add to the dynamic of the scene. The camera captures her upper body and face in a stationary, medium close-up shot..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001529_z.mp4
| |
000706_x
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man sits on a brown couch in a cozy living room, wearing a colorful sweater and striped pants. He energetically gestures toward a cartoonish poster on the wall, featuring a green hill, a tree, and clouds, with the word 'Generations' written at the top. As he talks, he shifts his posture and moves his hands, emphasizing his points. The camera captures the scene with a medium shot, showing both the man and the poster, while the background remains static..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000706_x.mp4
| |
001546_c
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman opens a window, inhales the fresh air with a smile, and then closes it gently after a few moments..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001546_c.mp4
| |
003482_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man wearing a red and white Arsenal jersey sits indoors, animatedly discussing a topic related to football. His hand gestures emphasize his points, and his facial expressions shift with his conversation. Behind him, a brick wall is decorated with Arsenal memorabilia, adding to the sports-themed atmosphere. The camera captures him from a medium shot, focusing on his upper body and face..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/003482_q.mp4
| |
000081_a
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A middle-aged man with short hair and a beard, dressed in a dark blue suit and white shirt, speaks directly to the camera. Set against a solid blue background, his serious and focused expression reflects a professional setting. The camera maintains a sharp focus on his face with even lighting, highlighting his features as he occasionally moves his head..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000081_a.mp4
| |
001509_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An adult bottlenose dolphin in a lab. Five scientists are walking around the dolphin, and some instruments are in the corner of the lab. A trainer squad is beside the dolphin to calm it..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001509_z.mp4
| |
001444_i
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
When winter finally arrived, bringing with it biting winds and a blanket of snow, Gus found himself hungry and shivering..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 4; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001444_i.mp4
| |
000905_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An old fishing village by the Zhuoshui River. Old fishing boats are moored on the riverbank, shrouded in thick fog..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000905_u.mp4
| |
000968_i
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Birds are flying over a forest with dried leaves and parched ground under the scorching midday sun, with a clear blue sky. Summer, Miyazaki comic style..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000968_i.mp4
| |
000206_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman with long, wavy brown hair and red lipstick sits in a modern, well-lit room. She wears a black floral dress and holds a white and red handbag with a heart-shaped design. Animatedly talking and gesturing, she occasionally points to the handbag, which features a black strap and silver hardware. The background includes a couch, a vase with pink flowers, and a bookshelf. The minimalist room has soft, neutral tones and clean lines..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000206_q.mp4
| |
000824_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Two women are sitting on the floor in a brightly lit room, laughing and enjoying a lively conversation while looking at a smartphone. One woman, with long brown hair, points at the phone screen, while the other, with blonde hair in a ponytail, reacts with amusement. The room has a playful vibe, with a colorful toy structure and blue wall in the background. Both are dressed casually, with the woman on the left wearing a white t-shirt and the woman on the right in a black 'LOVE & HOPE' sweatshirt. Their expressions show they are having fun with the content..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 4; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000824_u.mp4
| |
004781_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A child is preparing for an exam at night in cold weather. His mother is giving him a cup of tea, and the child is feeling tired..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004781_z.mp4
| |
001599_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man picks up a towel, wipes his face, and then briskly rubs his hands together to warm them up..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001599_m.mp4
| |
000710_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A college sign reads: 'Welcome to Harvard: Excellence and Innovation' as students walk across the campus, chatting with friends on their way to class..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000710_u.mp4
| |
004724_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An otherworldly landscape unfolds before us, with towering mountains that seem to touch the sky, their peaks crowned with swirling clouds of vibrant colors, like a dream painted by Andrei Platounov..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/004724_y.mp4
| |
001057_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A Viking woman moves away from the camera with slow steps. She walks slowly towards the darkness and briefly looks over her shoulder back at the camera. As she moves away from the camera, she slowly moves from right to left. The background is very dark, almost only mist is visible..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001057_m.mp4
| |
001453_v
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A basketball bounces on the gym floor, rolls towards the hoop, then swishes through the net as the crowd cheers..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001453_v.mp4
| |
001043_k
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A female scientist in a lab coat in a laboratory facing the lab bench, working. Then, her profile is visible, showing her handling a slide with FFPE tissue on it. Tilt down..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 1; physical/common-sense consistency: 1</answer>
|
./vs2_videos/001043_k.mp4
| |
004728_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man in a colorful floral shirt and white t-shirt is holding a white ceramic mug with blue patterns. He stands in a simple indoor setting, looking down at the mug, then glancing up at a hand reaching towards him, as if offering something. His expression shifts from contemplative to slightly surprised. The background features neutral-colored walls and a visible door. The camera captures the moment in a stationary medium close-up, focusing on the man's reactions and the gentle gesture of the hand..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004728_b.mp4
| |
002647_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man named 'LIGHTNING' sits in a green armchair in front of floral wallpaper. He wears a blue jacket over a gray shirt and has a name tag on his chest. The room feels casual and cozy, as he shifts his gaze and subtly changes his facial expressions, suggesting he's engaged in conversation or an interview. The camera captures him from the chest up with a steady, focused shot..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/002647_b.mp4
| |
004689_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Two Chinese women engage in a brutal battle in the midst of a desert enveloped in swirling yellow sands. One is clad in a white suit, wielding a sword, while the other wears a red suit and wields a shield. Their bodies are covered in blood..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 1; physical/common-sense consistency: 1</answer>
|
./vs2_videos/004689_q.mp4
| |
002608_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A swimmer is performing breaststroke, with the timer on the side of the pool showing 'Lap Time: 1 Minute, 10 Seconds'..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 2; physical/common-sense consistency: 1</answer>
|
./vs2_videos/002608_z.mp4
| |
004559_z
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A young woman stands at a wooden table in a sunny, tropical setting, holding a plastic bag of groceries. She wears a black 'Hollywood' t-shirt and blue jeans. In the background, another woman in a red shirt bends over, holding a green bucket. Palm trees and lush greenery surround them, and a colorful floral tablecloth covers the table. A black backpack rests on the table, with scattered leaves on the ground. The scene is captured from a slightly elevated angle, showing both women and the vibrant environment..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 4; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004559_z.mp4
| |
000684_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Small cupid statue made out of bronze. It is shiny. Warm light in the blurred background of an antique shop..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000684_u.mp4
| |
001170_h
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A cute squirrel Pomeranian with big eyes, a round body, soft and yellow fur, sitting and smiling next to flowers and grass, with a blue sky and white clouds above his head, sunshine, and a breeze, in cartoon 3D style..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001170_h.mp4
| |
000244_e
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A playful puppy dashes in a circle, halts to sniff the ground, then spins excitedly while chasing its own tail..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000244_e.mp4
| |
004648_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A diverse group of villagers, each showing a different expression, works together to solve a complex puzzle in the center of their village..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004648_b.mp4
| |
004885_f
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
The girl chases the butterfly around the field. The butterfly lands on a flower, and the girl slowly approaches it, trying to catch it, but the butterfly flies away..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004885_f.mp4
| |
001594_d
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A Yorkshire Terrier running on green grass in a deep forest with flowers and butterflies, with sunlight shining through the green trees..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 5; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001594_d.mp4
| |
000744_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A girl in a bikini is sunbathing by the blue sea under the bright sunlight. Next to her is a sun umbrella, while her boyfriend admires other beautiful women..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000744_m.mp4
| |
004926_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Red-haired girl at sunset on the beach, in the style of film from the 70s, with 35mm distortion and grainy texture..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004926_q.mp4
| |
000127_e
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Picture the yellowed pages of the book as Oliver begins to read, transporting him to a distant land where a young musician, Seraphina, faces trials mirroring his own struggles..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000127_e.mp4
| |
004663_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman brushes her long, flowing hair, ties it into a neat bun, and then steps confidently out the door..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 4</answer>
|
./vs2_videos/004663_q.mp4
| |
001425_h
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Calico cat swimming in a blue sea with coral reefs, colorful fish surrounding the corals. The cat swats at the fish, with rays of light illuminating the deep sea background..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 1; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001425_h.mp4
| |
002674_f
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
create a video of death as an elegant, regal, darkly beautiful woman in voluminous dark robes blowing in the wind, walking regally through an ancient medieval Middle Eastern caravan bazaar..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 4; physical/common-sense consistency: 3</answer>
|
./vs2_videos/002674_f.mp4
| |
004564_f
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
The lakeshore is adorned with Japanese Cherry Trees, pink wildflowers, and tall grass, providing an idyllic backdrop for our scientific pursuit..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 4; physical/common-sense consistency: 4</answer>
|
./vs2_videos/004564_f.mp4
| |
004843_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Show realistic lettuce plant roots absorbing tiny nanoparticles within a hydroponic system, with the camera focused on the roots during the process..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 3; physical/common-sense consistency: 5</answer>
|
./vs2_videos/004843_y.mp4
| |
000235_v
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A close-up of a pink plastic basket filled with fresh green banana leaves tied with a yellow string, resting under a pile of small, dark-colored fish. The vibrant, fresh scene highlights the natural textures of the leaves and fish, with the leaves placed neatly in the center of the basket and the fish stacked above. The camera remains steady, focusing on the vivid contrast of the green leaves and shiny, dark fish..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000235_v.mp4
| |
000183_r
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A gorgeous girl wearing glossy clothes is standing on a road. She is breathing, and her body is moving with her rhythmic breathing..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000183_r.mp4
| |
004754_e
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A young woman in anime style looking at her phone, irritated at what she is seeing, shaking her head, and rolling her eyes..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004754_e.mp4
| |
004639_q
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man running towards a group of enemy people with a sword in hand, wearing a white shirt covered in blood and dark navy pants..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/004639_q.mp4
| |
001060_s
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A small girl standing in a boat in the silence of the sea, with fire blazing in the sky, all alone, showing hope and bravery..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 3; physical/common-sense consistency: 4</answer>
|
./vs2_videos/001060_s.mp4
| |
004910_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Two cute girls chatting happily, fair-skinned, wearing headphones and bright, beautiful clothes, with a colorful translucent screen next to them..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 5; physical/common-sense consistency: 5</answer>
|
./vs2_videos/004910_y.mp4
| |
000600_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Female desert scavenger standing in front of a huge building, with a massive garage metal door opening slowly, revealing a gloomy light inside the building..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000600_u.mp4
| |
004901_a
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A dimly lit room with flickering candles casting eerie shadows on the walls. The camera zooms in on the narrator, who stands calmly amidst the candlelight. Pan up..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 4; physical/common-sense consistency: 2</answer>
|
./vs2_videos/004901_a.mp4
| |
004700_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A brown terrier runs excitedly after a colorful frisbee, leaps into the air with its front paws reaching, and then snatches the frisbee mid-flight with a triumphant bark..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004700_b.mp4
| |
001542_t
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Video of a sunrise in the desert with the Nile River in the background and some palm trees and brush in the foreground. The point of view is from inside a hut, with the door frame visible..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/001542_t.mp4
| |
001590_d
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
The camera pans over a vast battlefield, with armies of Marathas and Mughals positioned on opposite sides. The tension in the air is palpable, and the soldiers stand ready for the impending clash..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001590_d.mp4
| |
004600_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Quick cuts of chaos and pandemonium in the city. People are running in different directions, their faces filled with terror. The camera moves rapidly, capturing the frenzied atmosphere of the city. The background is filled with smoke and flickering, enhancing the sense of danger and urgency..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004600_y.mp4
| |
001570_v
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Cityscape Sunrise Revelation: Generate a video of the city skyline at sunrise, with subtle details revealing the conclusion of a noir narrative. Use warm hues and soft lighting to evoke a sense of revelation. Pan right..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/001570_v.mp4
| |
004680_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A young woman with glasses picks up an old hardcover book from a wooden desk, flips through the pages with curiosity, and then places a colorful bookmark between two pages to save her spot..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 2; physical/common-sense consistency: 4</answer>
|
./vs2_videos/004680_y.mp4
| |
004838_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Capture a photo of a mysterious and enchanting abandoned cottage in the heart of a forest. The photo should focus on a dusty old chest filled with art supplies inside the cottage. Among the supplies, there should be a beautiful golden paintbrush with a sparkling handle, catching the sunlight and radiating an air of magic. The scene should evoke curiosity and wonder, leaving viewers intrigued by the story that might unfold. Tilt down..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004838_b.mp4
| |
000640_n
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman sweeps the floor with a broom, collects the dust into a dustpan, and then empties the dustpan into the trash bin..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 2; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000640_n.mp4
| |
001402_m
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Create a video in anime Studio Ghibli style of a sunrise on serene Japanese rural farmland with lush green fields, a cozy homestead, mountains in the distance, and various animals like goats, llamas, chickens, and a Shiba Inu dog roaming around..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001402_m.mp4
| |
000940_n
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A fluffy rabbit jumps through a sunlit meadow, pauses to sniff a flower, and then leaps over a small stream..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000940_n.mp4
| |
000849_c
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A woman is eating a plate of slightly spicy food. Her cheeks are slightly red, her eyes are wider than usual, and her posture of holding the food in her hand is somewhat stiff. A few small beads of sweat pop out on her head, and her mouth is slightly open, as if trying to endure the faint spiciness. Small chili peppers are dancing around..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 4</answer>
|
./vs2_videos/000849_c.mp4
| |
001501_t
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A group of friends laughs together, one of them tells a joke, and another dramatically pretends to be shocked, covering their mouth in surprise..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001501_t.mp4
| |
004614_e
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A girl model stands in the middle of a busy street in Paris, where pedestrians pass by, and she looks at the camera. The camera slowly zooms out, catching her gaze. Fireworks explode behind her..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/004614_e.mp4
| |
004958_e
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A man in a gray suit and a woman in a colorful floral dress are having a friendly conversation in a simple classroom. The man stands near a wooden desk, while the woman walks towards the window, looking outside. The room features wooden desks, a whiteboard, and soft, natural lighting. Their interaction is casual and the camera captures the scene from an eye-level perspective, highlighting both the man and woman. The atmosphere is relaxed, with the focus on their conversation..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004958_e.mp4
| |
004678_y
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Two young girls are engaged in a creative activity, painting a white bowl with small cups and brushes. One girl, in a pink shirt, dips her brush into the paint, while the other, wearing a blue floral dress, assists or observes. The vibrant floral backdrop adds a cheerful atmosphere, and their focused, deliberate movements show their enjoyment of the process. The camera captures their activity at eye level, with no noticeable movement..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 4; physical/common-sense consistency: 5</answer>
|
./vs2_videos/004678_y.mp4
| |
004659_a
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Atv rider wearing a yellow and purple helmet walks to a yellow Can-Am Renegade ATV, sits on it, and starts the engine..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004659_a.mp4
| |
000967_n
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A lively coffee shop scene featuring three people. A man in a black suit stands at the center, holding a paper with a coffee-related message. He faces another man in a white shirt and red pants, who is engaged in the interaction. In the background, a woman in a white shirt and sunglasses observes. The setting includes a cozy coffee-themed ambiance with a coffee machine and a couch. The camera captures the moment from a wide-angle view, emphasizing the relaxed and playful atmosphere of the interaction..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 2; physical/common-sense consistency: 3</answer>
|
./vs2_videos/000967_n.mp4
| |
000246_d
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
Green leaves float slowly in the air, while the hour hand turns rapidly. The edge of the clock emits white light, and small water droplets are suspended around..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 1; physical/common-sense consistency: 1</answer>
|
./vs2_videos/000246_d.mp4
| |
001091_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A chef chops carrots and onions, sautés them in a sizzling pan, and sprinkles a pinch of paprika..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 1; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/001091_b.mp4
| |
001557_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An old and dusty antique shop filled with mysterious artifacts. The camera pans through the cluttered shelves, revealing ancient relics and trinkets. Among them, the spotlight falls on a mesmerizing old pocket watch with intricate engravings..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 5; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001557_u.mp4
| |
001509_u
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
An adult bottlenose dolphin in a lab. Five scientists are walking around the dolphin, and some instruments are in the corner of the lab. A trainer squad is beside the dolphin to calm it..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 4; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001509_u.mp4
| |
004887_h
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A person with glasses, a yellow jacket and yellow boots, holding a katana, fighting a monster in an anime 90s style..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 3; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/004887_h.mp4
| |
001115_b
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A child lighting a diya lamp on a Diwali night, with a rangoli design in the foreground and fireworks in the sky..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 2; text-to-video alignment: 3; physical/common-sense consistency: 3</answer>
|
./vs2_videos/001115_b.mp4
| |
000210_r
|
You are an expert for evaluating and thinking about the quality of AI videos from diverse dimensions.
We would like to evaluate its quality from three dimensions: 'visual quality', 'text-to-video alignment' and 'physical consistency'. Below is the definition of each dimension:
(1) visual quality:
The dimension 'visual quality' cares about the video's visual and optical propertities, including 'resolution, overall clarity, local blurriness, smoothness, stability of brightness/contrast, distortion/misalignment, abrupt changes, and any other factors the affect the watching experience'. The keywords written by the annotators are also mostly derived from the above factors.
(2) text-to-video alignment:
The dimension 't2v_alignment' mainly assesses whether the generated video fully and accurately depicts the elements mentioned in the text prompt, such as characters, actions, animals, etc., as well as background, quantity, color, weather, and so on. So the keywords written by annotators sometimes only indicate the elements that are missing from the video.
(3) physical/common-sense consistency:
The dimension 'physical/common-sense consistency' mainly examines whether there are any violations of common sense, physical laws, or any other aspects in the video that appear strange or unnatural. Most of the keywords provided by annotators point out the specific abnormalities or inconsistencies they observed in the video.
Here we provide an AI video generated by text-to-video models and its text prompt:
A sparrow lands on a windowsill, taps its beak against the glass, and then takes off into the sky..
Based on the video content and the dimension definitions, please evaluate the video quality and give the quality score.
The quality score must be integars in the range of 1 - 5.
|
video
|
video eval
|
[] |
<answer>visual quality: 5; text-to-video alignment: 3; physical/common-sense consistency: 2</answer>
|
./vs2_videos/000210_r.mp4
|
End of preview. Expand
in Data Studio
No dataset card yet
- Downloads last month
- 25