import gradio as gr from transformers import AutoProcessor, AutoModelForCausalLM import spaces 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) model = AutoModelForCausalLM.from_pretrained('HuggingFaceM4/Florence-2-DocVQA', trust_remote_code=True).to("cuda").eval() processor = AutoProcessor.from_pretrained('HuggingFaceM4/Florence-2-DocVQA', trust_remote_code=True) DESCRIPTION = "# [Florence-2-DocVQA Demo](https://huggingface.co/HuggingFaceM4/Florence-2-DocVQA)" 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 = model processor = processor if text_input is None: prompt = task_prompt else: prompt = task_prompt + text_input inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda") 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-large'): image = Image.fromarray(image) # Convert NumPy array to PIL Image if task_prompt == 'Document Visual Question Answering': task_prompt = '' results = run_example(task_prompt, image) return results, None elif task_prompt == 'Caption': task_prompt = '' results = run_example(task_prompt, image) return results, None elif task_prompt == 'Detailed Caption': task_prompt = '' results = run_example(task_prompt, image) return results, None elif task_prompt == 'More Detailed Caption': task_prompt = '' results = run_example(task_prompt, image) return results, None elif task_prompt == 'Object Detection': task_prompt = '' results = run_example(task_prompt, image) fig = plot_bbox(image, results['']) return results, fig_to_pil(fig) elif task_prompt == 'Dense Region Caption': task_prompt = '' results = run_example(task_prompt, image) fig = plot_bbox(image, results['']) return results, fig_to_pil(fig) elif task_prompt == 'Region Proposal': task_prompt = '' results = run_example(task_prompt, image) fig = plot_bbox(image, results['']) return results, fig_to_pil(fig) elif task_prompt == 'Caption to Phrase Grounding': task_prompt = '' results = run_example(task_prompt, image, text_input) fig = plot_bbox(image, results['']) return results, fig_to_pil(fig) elif task_prompt == 'Referring Expression Segmentation': task_prompt = '' results = run_example(task_prompt, image, text_input) output_image = copy.deepcopy(image) output_image = draw_polygons(output_image, results[''], fill_mask=True) return results, output_image elif task_prompt == 'Region to Segmentation': task_prompt = '' results = run_example(task_prompt, image, text_input) output_image = copy.deepcopy(image) output_image = draw_polygons(output_image, results[''], fill_mask=True) return results, output_image elif task_prompt == 'Open Vocabulary Detection': task_prompt = '' results = run_example(task_prompt, image, text_input) bbox_results = convert_to_od_format(results['']) fig = plot_bbox(image, bbox_results) return results, fig_to_pil(fig) elif task_prompt == 'Region to Category': task_prompt = '' results = run_example(task_prompt, image, text_input) return results, None elif task_prompt == 'Region to Description': task_prompt = '' results = run_example(task_prompt, image, text_input) return results, None elif task_prompt == 'OCR': task_prompt = '' results = run_example(task_prompt, image) return results, None elif task_prompt == 'OCR with Region': task_prompt = '' results = run_example(task_prompt, image) output_image = copy.deepcopy(image) output_image = draw_ocr_bboxes(output_image, results['']) return results, output_image else: return "", None # Return empty string and None for unknown task prompts css = """ #output { height: 500px; overflow: auto; border: 1px solid #ccc; } """ with gr.Blocks(css=css) as demo: gr.Markdown(DESCRIPTION) with gr.Tab(label="Florence-2 Image Captioning"): with gr.Row(): with gr.Column(): input_img = gr.Image(label="Input Picture") task_prompt = gr.Dropdown(choices=[ 'Document Visual Question Answering', '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' ], label="Task Prompt", value= 'Document Visual Question Answering') text_input = gr.Textbox(label="Text Input (optional)") 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], [output_text, output_img]) demo.launch(debug=True)