Upload 11 files

.gitattributes

@@ -33,3 +33,10 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Data.png filter=lfs diff=lfs merge=lfs -text
+image-demo.png filter=lfs diff=lfs merge=lfs -text
+image1.jpg filter=lfs diff=lfs merge=lfs -text
+image2.jpg filter=lfs diff=lfs merge=lfs -text
+Model.png filter=lfs diff=lfs merge=lfs -text
+Poster.png filter=lfs diff=lfs merge=lfs -text
+Tasks.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,14 +1,70 @@
----
-title: Florence 2 Demo
-emoji: 🏆
-colorFrom: indigo
-colorTo: red
-sdk: gradio
-sdk_version: 5.39.0
-app_file: app.py
-pinned: false
-license: mit
-short_description: 'This is demo of Florence-2: Advancing a Unified Representati'
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Florence-2 Demo: Advancing a Unified Representation for a Variety of Vision Tasks
+
+This is a Gradio-based demo showcasing **Florence-2**, a unified vision foundation model that advances the state-of-the-art in various computer vision tasks through a single, versatile architecture.
+
+## Demo Preview
+
+![Demo Screenshot](./image-demo.png)
+
+## About Florence-2
+
+Florence-2 represents a significant breakthrough in computer vision by providing a unified representation that can handle a diverse range of vision tasks including:
+
+- Object detection
+- Image captioning
+- Visual question answering
+- OCR (Optical Character Recognition)
+- Region proposal
+- Segmentation
+- And many more vision tasks
+
+The model demonstrates how a single architecture can be effectively applied across multiple vision domains, eliminating the need for task-specific models.
+
+## Paper & Resources
+
+📄 **CVPR 2024 Paper**: [Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks](https://openaccess.thecvf.com/content/CVPR2024/papers/Xiao_Florence-2_Advancing_a_Unified_Representation_for_a_Variety_of_Vision_CVPR_2024_paper.pdf)
+
+🎥 **CVPR Virtual Presentation**: [https://cvpr.thecvf.com/virtual/2024/poster/30529](https://cvpr.thecvf.com/virtual/2024/poster/30529)
+
+🖼️ **Research Poster**: [Poster.png](./Poster.png)
+
+## Demo Features
+
+This Gradio demo allows you to:
+- Upload images and interact with Florence-2's various capabilities
+- Test different vision tasks on your own images
+- Experience the unified model's performance across multiple domains
+
+## Getting Started
+
+1. Install the required dependencies:
+   ```bash
+   pip install -r requirements.txt
+   ```
+
+2. Run the demo:
+   ```bash
+   python app.py
+   ```
+
+3. Open your browser and navigate to the provided local URL to start using the demo.
+
+## References
+
+**Hugging Face Spaces**:
+- [Florence-2 Demo by gokaygokay](https://huggingface.co/spaces/gokaygokay/Florence-2)
+- [Florence-SAM Integration by SkalskiP](https://huggingface.co/spaces/SkalskiP/florence-sam)
+
+## Citation
+
+If you use this demo or find Florence-2 useful in your research, please cite:
+
+```bibtex
+@inproceedings{xiao2024florence,
+  title={Florence-2: Advancing a unified representation for a variety of vision tasks},
+  author={Xiao, Bin and Wu, Haiping and Xu, Weijian and Dai, Xiyang and Hu, Houdong and Lu, Yumao and Zeng, Michael and Liu, Ce and Yuan, Lu},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={4818--4829},
+  year={2024}
+}
+``` 

app.py

@@ -0,0 +1,290 @@
+import gradio as gr
+from transformers import AutoProcessor, AutoModelForCausalLM
+import spaces
+import torch
+import requests
+import copy
+from PIL import Image, ImageDraw, ImageFont 
+import io
+import matplotlib.pyplot as plt
+import matplotlib.patches as patches
+import random
+import numpy as np
+
+# import subprocess
+# subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+
+models = {
+    # 'microsoft/Florence-2-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).to(device).eval(),
+    # 'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to(device).eval(),
+    'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).to(device).eval(),
+    'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained("microsoft/Florence-2-base", torch_dtype=torch_dtype, trust_remote_code=True).to(device) # AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to(device).eval(),
+}
+
+processors = {
+    # 'microsoft/Florence-2-large-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True),
+    # 'microsoft/Florence-2-large': AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True),
+    'microsoft/Florence-2-base-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True),
+    'microsoft/Florence-2-base': AutoProcessor.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True) # AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
+}
+
+
+# Updated DESCRIPTION with the title
+DESCRIPTION = """
+# [Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks](https://arxiv.org/pdf/2311.06242)
+""" # [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-large)
+
+colormap = ['blue','orange','green','purple','brown','pink','gray','olive','cyan','red',
+            'lime','indigo','violet','aqua','magenta','coral','gold','tan','skyblue']
+
+def fig_to_pil(fig):
+    buf = io.BytesIO()
+    fig.savefig(buf, format='png')
+    buf.seek(0)
+    return Image.open(buf)
+
+@spaces.GPU
+def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florence-2-base'):
+    model = models[model_id]
+    processor = processors[model_id]
+    if text_input is None:
+        prompt = task_prompt
+    else:
+        prompt = task_prompt + text_input
+    inputs = processor(text=prompt, images=image, return_tensors="pt").to(device)
+    generated_ids = model.generate(
+        input_ids=inputs["input_ids"],
+        pixel_values=inputs["pixel_values"],
+        max_new_tokens=1024,
+        early_stopping=False,
+        do_sample=False,
+        num_beams=3,
+    )
+    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
+    parsed_answer = processor.post_process_generation(
+        generated_text,
+        task=task_prompt,
+        image_size=(image.width, image.height)
+    )
+    return parsed_answer
+
+def plot_bbox(image, data):
+    fig, ax = plt.subplots()
+    ax.imshow(image)
+    for bbox, label in zip(data['bboxes'], data['labels']):
+        x1, y1, x2, y2 = bbox
+        rect = patches.Rectangle((x1, y1), x2-x1, y2-y1, linewidth=1, edgecolor='r', facecolor='none')
+        ax.add_patch(rect)
+        plt.text(x1, y1, label, color='white', fontsize=8, bbox=dict(facecolor='red', alpha=0.5))
+    ax.axis('off')
+    return fig
+
+def draw_polygons(image, prediction, fill_mask=False):
+    draw = ImageDraw.Draw(image)
+    scale = 1
+    for polygons, label in zip(prediction['polygons'], prediction['labels']):
+        color = random.choice(colormap)
+        fill_color = random.choice(colormap) if fill_mask else None
+        for _polygon in polygons:
+            _polygon = np.array(_polygon).reshape(-1, 2)
+            if len(_polygon) < 3:
+                print('Invalid polygon:', _polygon)
+                continue
+            _polygon = (_polygon * scale).reshape(-1).tolist()
+            if fill_mask:
+                draw.polygon(_polygon, outline=color, fill=fill_color)
+            else:
+                draw.polygon(_polygon, outline=color)
+            draw.text((_polygon[0] + 8, _polygon[1] + 2), label, fill=color)
+    return image
+
+def convert_to_od_format(data):
+    bboxes = data.get('bboxes', [])
+    labels = data.get('bboxes_labels', [])
+    od_results = {
+        'bboxes': bboxes,
+        'labels': labels
+    }
+    return od_results
+
+def draw_ocr_bboxes(image, prediction):
+    scale = 1
+    draw = ImageDraw.Draw(image)
+    bboxes, labels = prediction['quad_boxes'], prediction['labels']
+    for box, label in zip(bboxes, labels):
+        color = random.choice(colormap)
+        new_box = (np.array(box) * scale).tolist()
+        draw.polygon(new_box, width=3, outline=color)
+        draw.text((new_box[0]+8, new_box[1]+2),
+                  "{}".format(label),
+                  align="right",
+                  fill=color)
+    return image
+
+def process_image(image, task_prompt, text_input=None, model_id='microsoft/Florence-2-base'):
+    image = Image.fromarray(image)  # Convert NumPy array to PIL Image
+    if task_prompt == 'Caption':
+        task_prompt = '<CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        return results, None
+    elif task_prompt == 'Detailed Caption':
+        task_prompt = '<DETAILED_CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        return results, None
+    elif task_prompt == 'More Detailed Caption':
+        task_prompt = '<MORE_DETAILED_CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        return results, None
+    elif task_prompt == 'Caption + Grounding':
+        task_prompt = '<CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        text_input = results[task_prompt]
+        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        results['<CAPTION>'] = text_input
+        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Detailed Caption + Grounding':
+        task_prompt = '<DETAILED_CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        text_input = results[task_prompt]
+        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        results['<DETAILED_CAPTION>'] = text_input
+        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'More Detailed Caption + Grounding':
+        task_prompt = '<MORE_DETAILED_CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        text_input = results[task_prompt]
+        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        results['<MORE_DETAILED_CAPTION>'] = text_input
+        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Object Detection':
+        task_prompt = '<OD>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        fig = plot_bbox(image, results['<OD>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Dense Region Caption':
+        task_prompt = '<DENSE_REGION_CAPTION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        fig = plot_bbox(image, results['<DENSE_REGION_CAPTION>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Region Proposal':
+        task_prompt = '<REGION_PROPOSAL>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        fig = plot_bbox(image, results['<REGION_PROPOSAL>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Caption to Phrase Grounding':
+        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Referring Expression Segmentation':
+        task_prompt = '<REFERRING_EXPRESSION_SEGMENTATION>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        output_image = copy.deepcopy(image)
+        output_image = draw_polygons(output_image, results['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True)
+        return results, output_image
+    elif task_prompt == 'Region to Segmentation':
+        task_prompt = '<REGION_TO_SEGMENTATION>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        output_image = copy.deepcopy(image)
+        output_image = draw_polygons(output_image, results['<REGION_TO_SEGMENTATION>'], fill_mask=True)
+        return results, output_image
+    elif task_prompt == 'Open Vocabulary Detection':
+        task_prompt = '<OPEN_VOCABULARY_DETECTION>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        bbox_results = convert_to_od_format(results['<OPEN_VOCABULARY_DETECTION>'])
+        fig = plot_bbox(image, bbox_results)
+        return results, fig_to_pil(fig)
+    elif task_prompt == 'Region to Category':
+        task_prompt = '<REGION_TO_CATEGORY>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        return results, None
+    elif task_prompt == 'Region to Description':
+        task_prompt = '<REGION_TO_DESCRIPTION>'
+        results = run_example(task_prompt, image, text_input, model_id)
+        return results, None
+    elif task_prompt == 'OCR':
+        task_prompt = '<OCR>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        return results, None
+    elif task_prompt == 'OCR with Region':
+        task_prompt = '<OCR_WITH_REGION>'
+        results = run_example(task_prompt, image, model_id=model_id)
+        output_image = copy.deepcopy(image)
+        output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
+        return results, output_image
+    else:
+        return "", None
+
+css = """
+  #output {
+    height: 500px; 
+    overflow: auto; 
+    border: 1px solid #ccc; 
+  }
+"""
+
+single_task_list = [
+    'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
+    'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
+    'Referring Expression Segmentation', 'Region to Segmentation',
+    'Open Vocabulary Detection', 'Region to Category', 'Region to Description',
+    'OCR', 'OCR with Region'
+]
+
+cascased_task_list = [
+    'Caption + Grounding', 'Detailed Caption + Grounding', 'More Detailed Caption + Grounding'
+]
+
+def update_task_dropdown(choice):
+    if choice == 'Cascased task':
+        return gr.Dropdown(choices=cascased_task_list, value='Caption + Grounding')
+    else:
+        return gr.Dropdown(choices=single_task_list, value='Caption')
+
+with gr.Blocks(css=css) as demo:
+    gr.Markdown(DESCRIPTION)
+    with gr.Row():
+        with gr.Column():
+            # Add overview.png image display
+            gr.Image("Tasks.png", label="Task Overview Image", type="pil", width=444, height=500)
+        with gr.Column():
+            gr.Image("Data.png", label="Data Engine", type="pil", elem_id="model_image", height=168)
+            gr.Image("Model.png", label="Model", type="pil", elem_id="model_image", height=316)
+    with gr.Tab(label="Florence-2 Demo"):
+        with gr.Row():
+            with gr.Column():
+                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-base')
+                task_type = gr.Radio(choices=['Single task', 'Cascased task'], label='Task type selector', value='Single task')
+                task_prompt = gr.Dropdown(choices=single_task_list, label="Task Prompt", value="Caption")
+                task_type.change(fn=update_task_dropdown, inputs=task_type, outputs=task_prompt)
+                text_input = gr.Textbox(label="Text Input (optional)")
+                input_img = gr.Image(label="Input Picture")
+                submit_btn = gr.Button(value="Submit")
+            with gr.Column():
+                output_text = gr.Textbox(label="Output Text")
+                output_img = gr.Image(label="Output Image")
+
+        gr.Examples(
+            examples=[
+                ["image1.jpg", 'Object Detection'],
+                ["image2.jpg", 'OCR with Region']
+            ],
+            inputs=[input_img, task_prompt],
+            outputs=[output_text, output_img],
+            fn=process_image,
+            cache_examples=True,
+            label='Try examples'
+        )
+
+        submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text, output_img])
+
+demo.launch(debug=True)

requirements.txt

@@ -0,0 +1,10 @@
+spaces
+timm
+flask==2.3.3
+werkzeug==2.3.7
+flask-cors==4.0.0
+torch>=2.0.0
+transformers>=4.30.0
+Pillow>=9.0.0
+matplotlib>=3.5.0
+numpy>=1.21.0