app.py

import os
import gradio as gr
from transformers import AutoModelForImageSegmentation
import torch
from torchvision import transforms
from PIL import Image

# --- 1. 初始化模型 ---
model_id = "briaai/RMBG-2.0"
print(f"正在載入模型: {model_id} ...")

hf_token = os.getenv("HF_TOKEN")

if not hf_token:
    print("⚠️ 警告: 未偵測到 HF_TOKEN，請檢查 Settings 中的 Secret。")

try:
    # 載入模型
    model = AutoModelForImageSegmentation.from_pretrained(
        model_id, 
        trust_remote_code=True, 
        token=hf_token
    )
    # 強制使用 CPU
    device = torch.device("cpu") 
    model.to(device)
    model.eval()
    print("✅ 模型載入成功！")
except Exception as e:
    print(f"❌ 模型載入失敗: {e}")

# --- 2. 定義圖像處理邏輯 ---
def process_image(input_image):
    if input_image is None:
        return None
    
    # 準備圖像尺寸
    image_size = (1024, 1024)
    
    # 定義轉換
    transform_image = transforms.Compose([
        transforms.Resize(image_size),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ])
    
    # 轉換並增加維度
    input_images = transform_image(input_image).unsqueeze(0).to(device)
    
    # 開始預測
    with torch.no_grad():
        preds = model(input_images)[-1].sigmoid().cpu()
    
    # 處理預測結果
    pred = preds[0].squeeze()
    
    # 轉回 PIL 圖片
    pred_pil = transforms.ToPILImage()(pred)
    
    # 調整回原始圖片的大小
    mask = pred_pil.resize(input_image.size)
    
    # 合成去背圖
    image = input_image.convert("RGBA")
    image.putalpha(mask)
    
    return image

# --- 3. 建立介面 (移除所有可能導致錯誤的外觀設定) ---
# 這裡去掉了 theme=... 參數，這是最安全的寫法
with gr.Blocks() as app:
    
    gr.Markdown("## ✂️ AI 自動去背 (RMBG 2.0)")
    
    with gr.Column():
        input_img = gr.Image(type="pil", label="上傳圖片", source="upload")
        btn = gr.Button("開始去背")
        output_img = gr.Image(type="pil", label="去背結果")

    btn.click(fn=process_image, inputs=input_img, outputs=output_img)

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