api/app.py

from fastapi import FastAPI, File, UploadFile
from fastapi.responses import Response
from PIL import Image
import io
import sys
import os
import torch
from safetensors.torch import load_file

# 添加src目录到系统路径
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))

from src.pipeline.pipeline import TattooRemovalPipeline
from src.model import UNet  # 假设您的模型类名为UNet

app = FastAPI()

def load_model():
    model = UNet()
    # 从huggingface下载模型权重
    weights_path = "tattoo_remover.safetensors"
    if not os.path.exists(weights_path):
        print("请先下载模型权重文件")
        print("从 https://huggingface.co/erickillian/tattoo-removal/blob/main/tattoo_remover.safetensors 下载")
        print("并放置在当前目录")
        return None
    
    state_dict = load_file(weights_path)
    model.load_state_dict(state_dict)
    return model

# 加载模型
model = load_model()
if model is not None:
    tattoo_remover = TattooRemovalPipeline(model=model)
else:
    print("模型加载失败")

@app.post("/remove-tattoo")
async def remove_tattoo(file: UploadFile = File(...)):
    if model is None:
        return {"error": "模型未正确加载"}
        
    # 读取上传的图片
    image_data = await file.read()
    image = Image.open(io.BytesIO(image_data)).convert('RGB')
    
    # 处理图片
    result = tattoo_remover(image)
    
    # 转换结果为字节流
    output = io.BytesIO()
    result.save(output, format='PNG')
    output.seek(0)
    
    return Response(content=output.getvalue(), media_type="image/png")

@app.get("/")
async def root():
    return {"message": "纹身移除API服务正在运行"}