app.py

import io
import torch
import gradio as gr
import numpy as np
from PIL import Image
from unimernet.common.config import Config
from unimernet.processors import load_processor
import unimernet.tasks as tasks
import argparse
import os

MAX_WIDTH = 872
MAX_HEIGHT = 1024

class ImageProcessor:
    """ImageProcessor handles model loading and image processing."""
    def __init__(self, cfg_path):
        self.cfg_path = cfg_path
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model, self.vis_processor = self.load_model_and_processor()

    def load_model_and_processor(self):
        args = argparse.Namespace(cfg_path=self.cfg_path, options=None)
        cfg = Config(args)
        task = tasks.setup_task(cfg)
        model = task.build_model(cfg).to(self.device)
        vis_processor = load_processor(
            "formula_image_eval",
            cfg.config.datasets.formula_rec_eval.vis_processor.eval,
        )
        return model, vis_processor

    def process_single_image(self, pil_image):
        image = self.vis_processor(pil_image).unsqueeze(0).to(self.device)
        output = self.model.generate({"image": image})
        pred = output["pred_str"][0]
        return pred

# 初始化模型
cfg_path = "demo.yaml"
processor = ImageProcessor(cfg_path)

# 单张预测
def predict_single(img):
    if img is None:
        return "No image uploaded."
    img = img.convert("RGB")
    img.thumbnail((MAX_WIDTH, MAX_HEIGHT), Image.Resampling.LANCZOS)
    latex_code = processor.process_single_image(img)
    return latex_code

# 批量预测
def predict_batch(img_list):
    if not img_list:
        return ["No images uploaded."]
    results = []
    for img in img_list:
        if img is None:
            results.append("Invalid image")
            continue
        img = img.convert("RGB")
        img.thumbnail((MAX_WIDTH, MAX_HEIGHT), Image.Resampling.LANCZOS)
        latex_code = processor.process_single_image(img)
        results.append(latex_code)
    return results

# 界面搭建
title = "UniMERNet Formula Recognition"
description = "Upload an image (or multiple images) containing math formulas. The model will return LaTeX code."

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"# {title}")
    gr.Markdown(description)

    with gr.Tab("Single Image Recognition"):
        with gr.Row():
            input_img = gr.Image(type="pil", label="Upload a single formula image")
            output_text = gr.Textbox(label="Predicted LaTeX code", lines=5)

        btn_single = gr.Button("Recognize Single Image")
        btn_single.click(fn=predict_single, inputs=input_img, outputs=output_text)

    # with gr.Tab("Batch Image Recognition"):
    #     with gr.Row():
    #         input_imgs = gr.File(file_types=["image"], file_count="multiple", label="Upload multiple images (png/jpg/jpeg/webp)")
    #
    #     batch_outputs = gr.Dataframe(headers=["Image", "Predicted LaTeX Code"], datatype=["str", "str"])
    #
    #     def batch_process(files):
    #         imgs = []
    #         file_names = []
    #         for file in files:
    #             with Image.open(file.name) as img:
    #                 imgs.append(img.copy())
    #                 file_names.append(os.path.basename(file.name))
    #         preds = predict_batch(imgs)
    #         return list(zip(file_names, preds))
    #
    #     btn_batch = gr.Button("Recognize Batch Images")
    #     btn_batch.click(fn=batch_process, inputs=input_imgs, outputs=batch_outputs)
    with gr.Tab("Batch Image Recognition"):
        with gr.Row():
            input_imgs = gr.File(file_types=["image"], file_count="multiple",
                                 label="Upload multiple images (png/jpg/jpeg/webp)")

        batch_outputs = gr.Dataframe()  # ✅ 改这里，不加headers和datatype！


        def batch_process(files):
            imgs = []
            file_names = []
            for file in files:
                with Image.open(file.name) as img:
                    imgs.append(img.copy())
                    file_names.append(os.path.basename(file.name))
            preds = predict_batch(imgs)
            return list(zip(file_names, preds))


        btn_batch = gr.Button("Recognize Batch Images")
        btn_batch.click(fn=batch_process, inputs=input_imgs, outputs=batch_outputs)

if __name__ == "__main__":
    demo.launch()