app.py

import os
import gradio as gr
import huggingface_hub
import numpy as np
import onnxruntime as rt
import pandas as pd
from PIL import Image
from huggingface_hub import login

# 模型配置
MODEL_REPO = "SmilingWolf/wd-swinv2-tagger-v3"  # 默认模型
MODEL_FILENAME = "model.onnx"
LABEL_FILENAME = "selected_tags.csv"
HF_TOKEN = os.environ.get("HF_TOKEN", "")

if not os.environ.get("HF_TOKEN"):
    print("⚠️ 警告：未检测到HF_TOKEN，部分模型可能需要认证")
else:
    login(token=os.environ.get("HF_TOKEN"))

# 标签处理配置
kaomojis = [
    "0_0",
    "(o)_(o)",
    "+_+",
    "+_-",
    "._.",
    "<o>_<o>",
    "<|>_<|>",
    "=_=",
    ">_<",
    "3_3",
    "6_9",
    ">_o",
    "@_@",
    "^_^",
    "o_o",
    "u_u",
    "x_x",
    "|_|",
    "||_||",
]

class Tagger:
    def __init__(self):
        self.model = None
        self.tag_names = []
        self.model_size = None
        self.hf_token = os.environ.get("HF_TOKEN", "")  # 从环境变量获取
        self._init_model()

    def _init_model(self):
        """初始化模型和标签"""
        try:
            label_path = huggingface_hub.hf_hub_download(
                MODEL_REPO, 
                LABEL_FILENAME,
                token=self.hf_token
            )
            model_path = huggingface_hub.hf_hub_download(
                MODEL_REPO,
                MODEL_FILENAME,
                token=self.hf_token
            )
            
            # 加载标签
            tags_df = pd.read_csv(label_path)
            self.tag_names = tags_df["name"].tolist()
            self.categories = {
                "rating": np.where(tags_df["category"] == 9)[0],
                "general": np.where(tags_df["category"] == 0)[0],
                "character": np.where(tags_df["category"] == 4)[0]
            }
            
            # 加载ONNX模型
            self.model = rt.InferenceSession(model_path)
            self.model_size = self.model.get_inputs()[0].shape[1]
        except huggingface_hub.utils.HfHubHTTPError as e:
            if "401" in str(e):
                raise RuntimeError(
                    "模型下载认证失败，请：\n"
                    "1. 访问https://huggingface.co/SmilingWolf/wd-swinv2-tagger-v3\n"
                    "2. 点击Agree and continue\n"
                    "3. 确保HF_TOKEN已正确设置"
                )
            else:
                raise

    def _preprocess(self, img):
        """图像预处理"""
        # 转换为RGB
        if img.mode != "RGB":
            img = img.convert("RGB")
        
        # 填充为正方形
        size = max(img.size)
        padded = Image.new("RGB", (size, size), (255, 255, 255))
        padded.paste(img, ((size - img.width)//2, (size - img.height)//2))
        
        # 调整尺寸
        if size != self.model_size:
            padded = padded.resize((self.model_size, self.model_size), Image.BICUBIC)
        
        # 转换为BGR格式
        return np.array(padded)[:, :, ::-1].astype(np.float32)

    def predict(self, img, general_thresh=0.35, character_thresh=0.85):
        """执行预测"""
        # 预处理
        img_data = self._preprocess(img)[np.newaxis]
        
        # 运行模型
        input_name = self.model.get_inputs()[0].name
        outputs = self.model.run(None, {input_name: img_data})[0][0]
        
        # 组织结果
        results = {
            "ratings": {},
            "general": {},
            "characters": {}
        }
        
        # 处理评分标签
        for idx in self.categories["rating"]:
            tag = self.tag_names[idx].replace("_", " ")
            results["ratings"][tag] = float(outputs[idx])
        
        # 处理通用标签
        for idx in self.categories["general"]:
            if outputs[idx] > general_thresh:
                tag = self.tag_names[idx].replace("_", " ")
                results["general"][tag] = float(outputs[idx])
        
        # 处理角色标签
        for idx in self.categories["character"]:
            if outputs[idx] > character_thresh:
                tag = self.tag_names[idx].replace("_", " ")
                results["characters"][tag] = float(outputs[idx])
        
        # 排序结果
        results["general"] = dict(sorted(
            results["general"].items(),
            key=lambda x: x[1],
            reverse=True
        ))
        
        return results

# 创建Gradio界面
with gr.Blocks(theme=gr.themes.Soft(), title="AI图像标签分析器") as demo:
    gr.Markdown("# 🖼️ AI图像标签分析器")
    gr.Markdown("上传图片自动分析图像内容标签")
    
    with gr.Row():
        with gr.Column(scale=1):
            img_input = gr.Image(type="pil", label="上传图片")
            with gr.Accordion("高级设置", open=False):
                general_slider = gr.Slider(0, 1, 0.35, 
                    label="通用标签阈值", 
                    info="值越高标签越少但更准确")
                char_slider = gr.Slider(0, 1, 0.85,
                    label="角色标签阈值",
                    info="推荐保持较高阈值")
            analyze_btn = gr.Button("开始分析", variant="primary")

        with gr.Column(scale=2):
            with gr.Tabs():
                with gr.TabItem("🏷️ 通用标签"):
                    general_tags = gr.Label(label="检测到的通用标签")
                with gr.TabItem("👤 角色标签"):
                    char_tags = gr.Label(label="检测到的角色标签")
                with gr.TabItem("⭐ 评分标签"):
                    rating_tags = gr.Label(label="图像评级标签")
            
            output_text = gr.Textbox(label="标签文本",
                placeholder="生成的标签文本将显示在这里...")

    # 处理逻辑
    def process_image(img, gen_thresh, char_thresh):
        tagger = Tagger()
        results = tagger.predict(img, gen_thresh, char_thresh)
        
        # 格式化文本输出
        tag_text = ", ".join(results["general"].keys())
        if results["characters"]:
            tag_text += ", " + ", ".join(results["characters"].keys())
        
        return {
            general_tags: results["general"],
            char_tags: results["characters"],
            rating_tags: results["ratings"],
            output_text: tag_text
        }

    analyze_btn.click(
        process_image,
        inputs=[img_input, general_slider, char_slider],
        outputs=[general_tags, char_tags, rating_tags, output_text]
    )

# 启动应用
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)