feifeilib/feifeiflorencebase.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-base": AutoModelForCausalLM.from_pretrained(
        "PJMixers-Images/Florence-2-base-gemini-2.0-flash-thinking-exp-1219-v0.2", trust_remote_code=True
    )
    .to("cuda")
    .eval()
}

processors = {
    "microsoft/Florence-2-base": AutoProcessor.from_pretrained(
        "PJMixers-Images/Florence-2-base-gemini-2.0-flash-thinking-exp-1219-v0.2", trust_remote_code=True
    )
}


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="<MORE_DETAILED_CAPTION>",
    image=None,
    text_input=None,
    model_id="microsoft/Florence-2-base",
    progress=gr.Progress(track_tqdm=True),
):
    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 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="More Detailed Caption",
    text_input=None,
    model_id="microsoft/Florence-2-base",
):
    image = Image.open(image).convert("RGB")
    base_height = 512
    h_percent = base_height / float(image.size[1])
    w_size = int((float(image.size[0]) * float(h_percent)))
    image = image.resize((w_size, base_height), Image.LANCZOS)

    if task_prompt == "Caption":
        task_prompt = "<CAPTION>"
        results = run_example(task_prompt, image, model_id=model_id)
        return results
    elif task_prompt == "Detailed Caption":
        task_prompt = "<DETAILED_CAPTION>"
        results = run_example(task_prompt, image, model_id=model_id)
        return results
    elif task_prompt == "More Detailed Caption":
        task_prompt = "<MORE_DETAILED_CAPTION>"
        results = run_example(task_prompt, image, model_id=model_id)
        results = results[task_prompt]
        return results
    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
    elif task_prompt == "Region to Description":
        task_prompt = "<REGION_TO_DESCRIPTION>"
        results = run_example(task_prompt, image, text_input, model_id)
        return results
    elif task_prompt == "OCR":
        task_prompt = "<OCR>"
        results = run_example(task_prompt, image, model_id=model_id)
        return results
    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  # Return empty string and None for unknown task prompts


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")