app.py

import os
import argparse
import numpy as np
from PIL import Image
import torch
import torchvision.transforms as T
from transformers import AutoTokenizer
import gradio as gr
from resnet50 import build_model
from utils import generate_similiarity_map, post_process, load_tokenizer, build_transform_R50
from utils import IMAGENET_MEAN, IMAGENET_STD
from internvl.train.dataset import dynamic_preprocess
from internvl.model.internvl_chat import InternVLChatModel
import spaces

# 模型配置
CHECKPOINTS = {
    "TokenFD_4096_English_seg": "TongkunGuan/TokenFD_4096_English_seg",
    "TokenFD_2048_Bilingual_seg": "TongkunGuan/TokenFD_2048_Bilingual_seg",
}

# 全局变量
HF_TOKEN = os.getenv("HF_TOKEN")

def load_model(check_type):
    # device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    device = torch.device("cuda")
    if check_type == 'R50':
        tokenizer = load_tokenizer('tokenizer_path')
        model = build_model(argparse.Namespace()).eval()
        model.load_state_dict(torch.load(CHECKPOINTS['R50'], map_location='cpu')['model'])
        transform = build_transform_R50(normalize_type='imagenet')

    elif check_type == 'R50_siglip':
        tokenizer = load_tokenizer('tokenizer_path')
        model = build_model(argparse.Namespace()).eval()
        model.load_state_dict(torch.load(CHECKPOINTS['R50_siglip'], map_location='cpu')['model'])
        transform = build_transform_R50(normalize_type='imagenet')

    elif 'TokenFD' in check_type:
        model_path = CHECKPOINTS[check_type]
        tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, use_fast=False, use_auth_token=HF_TOKEN)
        model = InternVLChatModel.from_pretrained(model_path, torch_dtype=torch.bfloat16).eval()
        transform = T.Compose([
            T.Lambda(lambda img: img.convert('RGB')),
            T.Resize((224, 224)),
            T.ToTensor(),
            T.Normalize(IMAGENET_MEAN, IMAGENET_STD)
        ])
    
    return model.to(device), tokenizer, transform, device

def process_image(model, tokenizer, transform, device, check_type, image, text, state):
    src_size = image.size
    if 'TokenOCR' in check_type:
        images, target_ratio = dynamic_preprocess(image, min_num=1, max_num=12, 
                                                  image_size=model.config.force_image_size,
                                                  use_thumbnail=model.config.use_thumbnail,
                                                  return_ratio=True)
        pixel_values = torch.stack([transform(img) for img in images]).to(device)
    else:
        pixel_values = torch.stack([transform(image)]).to(device)
        target_ratio = (1, 1)

    # 文本处理
    text += ' '
    input_ids = tokenizer(text)['input_ids'][1:]
    input_ids = torch.tensor(input_ids, device=device)
    
    # 获取嵌入
    with torch.no_grad():
        if 'R50' in check_type:
            text_embeds = model.language_embedding(input_ids)
        else:
            text_embeds = model.tok_embeddings(input_ids)
        
        vit_embeds, size1 = model.forward_tokenocr(pixel_values.to(torch.bfloat16).to(device))
        vit_embeds, size2 = post_process(vit_embeds, target_ratio, check_type)
        
        # 计算相似度
        text_embeds = text_embeds / text_embeds.norm(dim=-1, keepdim=True)
        vit_embeds = vit_embeds / vit_embeds.norm(dim=-1, keepdim=True)
        similarity = text_embeds @ vit_embeds.T
        resized_size = size1 if size1 is not None else size2

    attn_map = similarity.reshape(len(text_embeds), resized_size[0], resized_size[1])
    all_bpe_strings = [tokenizer.decode(input_id) for input_id in input_ids]
    vis = generate_similiarity_map([image], attn_map, 
                                   [tokenizer.decode([i]) for i in input_ids], 
                                   [], target_ratio, src_size)
    
    bpe = [tokenizer.decode([i]) for i in input_ids]
    bpe[-1] = text

    # Store results in state
    state['current_vis'] = vis
    state['current_bpe'] = bpe
    return image, vis[0], bpe[0], len(vis) - 1

# Gradio界面
with gr.Blocks() as demo:
    gr.Markdown("## BPE Visualization Demo - TokenFD基座模型能力可视化")
    
    with gr.Row():
        with gr.Column(scale=0.5):
            model_type = gr.Dropdown(
                choices=["TokenFD_4096_English_seg", "TokenFD_2048_Bilingual_seg", "R50", "R50_siglip"],
                label="Select model type",
                value="TokenOCR_4096_English_seg"
            )
            image_input = gr.Image(label="Upload images", type="pil")
            text_input = gr.Textbox(label="Input text")
            run_btn = gr.Button("RUN")
            gr.Examples(
                examples=[
                    [os.path.join("examples", "examples0.jpg"), "Veterans and Benefits"],
                    [os.path.join("examples", "examples1.jpg"), "Refreshers"],
                    [os.path.join("examples", "examples2.png"), "Vision Transformer"]
                ],
                inputs=[image_input, text_input],
                label="Sample input"
            )
        
        with gr.Column(scale=2):
            gr.Markdown("<p style='font-size:20px;'><span style='color:red;'>If the input text is not included in the image</span>, the attention map will show a lot of noise (the actual response value is very low), since we normalize the attention map according to the relative value.</p>")
            orig_img = gr.Image(label="Original picture", interactive=False)
            heatmap = gr.Image(label="BPE visualization", interactive=False)
            prev_btn = gr.Button("⬅ Last", visible=False)
            index_slider = gr.Slider(0, 1, value=0, step=1, label="BPE index", visible=False)
            next_btn = gr.Button("⮕ Next", visible=False)
            bpe_display = gr.Markdown("Current BPE: ", visible=False)

    state = gr.State()
    state['current_vis'] = []
    state['current_bpe'] = []
    state['current_index'] = 0

    @spaces.GPU
    def on_run_clicked(model_type, image, text, state):
        image, vis, bpe, slider_max_val = process_image(*load_model(model_type), model_type, image, text, state)
        state['current_vis'] = vis
        state['current_bpe'] = bpe
        state['current_index'] = 0
        bpe_text = format_bpe_display(bpe)
        return image, vis, bpe_text, slider_max_val

    run_btn.click(
        on_run_clicked,
        inputs=[model_type, image_input, text_input, state],
        outputs=[orig_img, heatmap, bpe_display],
        _js="""
            (orig_img, heatmap, bpe_display, slider_max_val) => {
                index_slider.update({ visible: true, maximum: slider_max_val, value: 0 });
                prev_btn.update({ visible: true });
                next_btn.update({ visible: true });
                return [orig_img, heatmap, bpe_display];
            }
        """
    )

    prev_btn.click(
        lambda state: update_index(-1, state),
        inputs=[state],
        outputs=[heatmap, bpe_display, index_slider]
    )

    next_btn.click(
        lambda state: update_index(1, state),
        inputs=[state],
        outputs=[heatmap, bpe_display, index_slider]
    )

    index_slider.change(
        lambda x, state: update_slider_index(x, state),
        inputs=[index_slider, state],
        outputs=[heatmap, bpe_display]
    )

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