Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -8,7 +8,9 @@ from typing import Dict
|
|
8 |
import gradio as gr
|
9 |
import ollama
|
10 |
from PIL import Image
|
|
|
11 |
|
|
|
12 |
|
13 |
def save_temp_image(image: Image.Image) -> str:
|
14 |
"""
|
@@ -27,45 +29,87 @@ def save_temp_image(image: Image.Image) -> str:
|
|
27 |
return tmp_file.name
|
28 |
|
29 |
def id_extractor(image: Image.Image) -> Dict:
|
30 |
-
"""
|
31 |
-
Extracts key details from the provided image using the ollama chat model.
|
32 |
|
33 |
-
|
34 |
-
|
35 |
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
# Define the Gradio interface for the ID extractor
|
71 |
id_interface = gr.Interface(
|
|
|
8 |
import gradio as gr
|
9 |
import ollama
|
10 |
from PIL import Image
|
11 |
+
from qwen_vl_utils import process_vision_info
|
12 |
|
13 |
+
from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
|
14 |
|
15 |
def save_temp_image(image: Image.Image) -> str:
|
16 |
"""
|
|
|
29 |
return tmp_file.name
|
30 |
|
31 |
def id_extractor(image: Image.Image) -> Dict:
|
|
|
|
|
32 |
|
33 |
+
|
34 |
+
# default: Load the model on the available device(s)
|
35 |
|
36 |
+
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
37 |
+
|
38 |
+
"Qwen/Qwen2.5-VL-7B-Instruct", torch_dtype="auto", device_map="auto"
|
39 |
+
|
40 |
+
)
|
41 |
+
|
42 |
+
# default processer
|
43 |
+
|
44 |
+
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")
|
45 |
+
|
46 |
+
messages = [
|
47 |
+
|
48 |
+
{
|
49 |
+
|
50 |
+
"role": "user",
|
51 |
+
|
52 |
+
"content": [
|
53 |
+
|
54 |
+
{
|
55 |
+
|
56 |
+
"type": "image",
|
57 |
+
|
58 |
+
"image": image,
|
59 |
+
|
60 |
+
},
|
61 |
+
|
62 |
+
{"type": "text", "text": "Extract key details like 'name', 'date of birth', 'ID number', 'Issuer' from the image as JSON, excluding signatures. Note if a ID has two names, pick the first one."},
|
63 |
+
|
64 |
+
],
|
65 |
+
|
66 |
+
}
|
67 |
+
|
68 |
+
]
|
69 |
+
|
70 |
+
# Preparation for inference
|
71 |
+
|
72 |
+
text = processor.apply_chat_template(
|
73 |
+
|
74 |
+
messages, tokenize=False, add_generation_prompt=True
|
75 |
+
|
76 |
+
)
|
77 |
+
|
78 |
+
image_inputs, video_inputs = process_vision_info(messages)
|
79 |
+
|
80 |
+
inputs = processor(
|
81 |
+
|
82 |
+
text=[text],
|
83 |
+
|
84 |
+
images=image_inputs,
|
85 |
+
|
86 |
+
videos=video_inputs,
|
87 |
+
|
88 |
+
padding=True,
|
89 |
+
|
90 |
+
return_tensors="pt",
|
91 |
+
|
92 |
+
)
|
93 |
+
|
94 |
+
|
95 |
+
# Inference: Generation of the output
|
96 |
+
|
97 |
+
generated_ids = model.generate(**inputs, max_new_tokens=128)
|
98 |
+
|
99 |
+
generated_ids_trimmed = [
|
100 |
+
|
101 |
+
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
102 |
+
|
103 |
+
]
|
104 |
+
|
105 |
+
output_text = processor.batch_decode(
|
106 |
+
|
107 |
+
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
108 |
+
|
109 |
+
)
|
110 |
+
|
111 |
+
return output_text
|
112 |
+
|
113 |
|
114 |
# Define the Gradio interface for the ID extractor
|
115 |
id_interface = gr.Interface(
|