app.py

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)

models = {
    'microsoft/Florence-2-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).to("cuda").eval(),
    'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to("cuda").eval(),
    'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).to("cuda").eval(),
    'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to("cuda").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),
}


DESCRIPTION = "# [Florence-2 OCR 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-large'):
    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("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 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 == 'OCR':
        task_prompt = '<OCR>'
        results = run_example(task_prompt, image, model_id=model_id)
        return results, None
    else:
        return "", None  # Return empty string and None for unknown task prompts

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.Tab(label="Florence-2 Image Captioning"):
        with gr.Row():
            with gr.Column():
                input_img = gr.Image(label="Input Picture")
                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-large')
                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)")
                submit_btn = gr.Button(value="Submit")
            with gr.Column():
                output_text = gr.Textbox(label="Output Text")
               

        gr.Examples(
            examples=[
                ["image1.jpg", 'Object Detection'],
                ["image2.jpg", 'OCR with Region']
            ],
            inputs=[input_img, task_prompt],
            outputs=[output_text],
            fn=process_image,
            cache_examples=True,
            label='Try examples'
        )

        submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text])

demo.launch(debug=True)