feat : deblur add new module

.gitignore

@@ -0,0 +1,27 @@
+# 1. 虛擬環境目錄
+.venv
+venv/
+env/
+/site-packages
+
+# 2. Python 編譯快取
+__pycache__/*
+*.pyc
+*.pyd
+*.so
+
+# 3. 測試和文件
+.pytest_cache/
+htmlcov/
+.coverage
+
+# 4. 編輯器/IDE 檔案 (可選，依據您使用的工具)
+.idea/          # PyCharm/IntelliJ 專案檔案
+.vscode/        # VS Code 設定 (如果不想共享)
+*.swp           # Vim 臨時檔案
+
+# 5. 您的靜態/媒體檔案 (保持原樣，但使用更精確的模式)
+/static/       # 忽略整個 static 資料夾
+/logs/         # 日誌檔案
+*.log
+*.sqlite3      # 如果使用 SQLite 資料庫

app.py

@@ -1,19 +1,17 @@
-from fastapi import FastAPI, Request, Response
+from fastapi import FastAPI, Request, Response, Form
 from fastapi.responses import JSONResponse
 from fastapi.staticfiles import StaticFiles
+from fastapi.middleware.cors import CORSMiddleware  # 匯入 FastAPI 的 CORS 中介軟體
 import requests
+from typing import Annotated # 推薦用於 Pydantic v2+
+
+from services.deblur import deblur_image_tiled
 
-import traceback
-import torch
-import torch.nn as nn
-from torchvision import transforms
 from PIL import Image
 import io
-import numpy as np
 import os
 from datetime import datetime
-
-from models.fpn_inception import FPNInception  # 你自己的模型類別
+import uvicorn
 
 STATIC_DIR = "static"
 
@@ -23,71 +21,20 @@ os.environ["TRANSFORMERS_CACHE"] = "./.cache"
 os.makedirs("./.cache", exist_ok=True)
 os.makedirs(STATIC_DIR, exist_ok=True)
 
-# =====================
-# 初始化模型
-# =====================
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-print(f"🔹 Using device: {device}")
-
-checkpoint_path = os.path.join("model", "deblurgan_v2_latest.pth")
-
-G = FPNInception(norm_layer=nn.InstanceNorm2d).to(device)
-
-try:
-    checkpoint = torch.load(checkpoint_path, map_location=device)
-except Exception as e:
-    print("❌ checkpoint 載入錯誤：", e)
-    traceback.print_exc()
-
-G.load_state_dict(checkpoint["G"], strict=False)
-G.eval()
-print("✅ Model loaded from", checkpoint_path)
-
-# =====================
-# Tile-based 推論函式
-# =====================
-def deblur_image_tiled(model, img, device, tile_size=512, overlap=32):
-    model.eval()
-    w, h = img.size
-    new_w = (w // 32) * 32
-    new_h = (h // 32) * 32
-    if new_w != w or new_h != h:
-        img = img.resize((new_w, new_h), Image.BICUBIC)
-        w, h = new_w, new_h
-
-    img_np = np.array(img).astype(np.float32) / 255.0
-    img_tensor = torch.from_numpy(img_np).permute(2, 0, 1).unsqueeze(0).to(device)
-
-    stride = tile_size - overlap
-    tiles_x = list(range(0, w, stride))
-    tiles_y = list(range(0, h, stride))
-    if tiles_x[-1] + tile_size > w:
-        tiles_x[-1] = w - tile_size
-    if tiles_y[-1] + tile_size > h:
-        tiles_y[-1] = h - tile_size
-
-    output = torch.zeros_like(img_tensor)
-    weight = torch.zeros_like(img_tensor)
-
-    with torch.no_grad():
-        for y in tiles_y:
-            for x in tiles_x:
-                patch = img_tensor[:, :, y:y+tile_size, x:x+tile_size]
-                pred = model(patch)
-                output[:, :, y:y+tile_size, x:x+tile_size] += pred
-                weight[:, :, y:y+tile_size, x:x+tile_size] += 1.0
-
-    output /= weight
-    output = torch.clamp(output, 0, 1)
-    out_np = (output.squeeze().permute(1, 2, 0).cpu().numpy() * 255.0).astype(np.uint8)
-    return Image.fromarray(out_np)
-
 # =====================
 # 初始化 FastAPI
 # =====================
 app = FastAPI(title="DeblurGANv2 API")
 
 app.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
+# 設定 CORS (跨來源資源共用)
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],  # 允許所有來源
+    allow_credentials=True,  # 允許憑證
+    allow_methods=["*"],  # 允許所有 HTTP 方法
+    allow_headers=["*"],  # 允許所有 HTTP 標頭
+)
 
 # =====================
 # API 路由
@@ -109,17 +56,19 @@ def greet_json(request: Request, response: Response):
     return JSONResponse(content={"message": "Hello World", "client": client_host})
 
 @app.post("/predict")
-async def predict( request: Request, response: Response ):
+async def predict(
+    request: Request,
+    # 將您的 form-data 欄位定義為函數參數，並使用 Form()
+    file_name: Annotated[str, Form(description="檔案名稱")],
+    file_format: Annotated[str, Form(description="檔案格式")],
+    file_url: Annotated[str, Form(description="檔案下載網址")], # 可選參數，有預設值
+    # 對於需要轉換類型 (例如 int) 的欄位，直接在類型提示中指定
+    file_width: Annotated[int, Form(description="檔案寬度")] = 0,
+    file_height: Annotated[int, Form(description="檔案高度")] = 0,
+    file_created_at: Annotated[str, Form(description="檔案建立時間")] = None,
+):
     try:
         print("### start /predict !!")
-        # 1️⃣ 讀取 form-data
-        form = await request.form()
-        file_name = form.get("file_name")
-        file_format = form.get("file_format")
-        file_url = form.get("file_url")
-        file_width = int(form.get("file_width", 0))
-        file_height = int(form.get("file_height", 0))
-        file_created_at = form.get("file_created_at")
 
         if not file_url:
             return JSONResponse(
@@ -133,7 +82,7 @@ async def predict( request: Request, response: Response ):
         img = Image.open(io.BytesIO(resp.content)).convert("RGB")
 
         # 3️⃣ 去模糊
-        result = deblur_image_tiled(G, img, device)
+        result = deblur_image_tiled(img)
 
         # 4️⃣ 產生檔名
         base_name = f"{file_name}_{file_width}_{file_height}_{file_created_at}.jpg"
@@ -159,4 +108,8 @@ async def predict( request: Request, response: Response ):
         traceback.print_exc()
         return JSONResponse(
             {"status": "error", "message": str(e)}, status_code=500
-        )
+        )
+    
+
+if __name__ == "__main__":
+    uvicorn.run("app:app", host="0.0.0.0", port=7860, reload=False)

requirements.txt

@@ -1,4 +1,4 @@
-fastapi
+fastapi[all]
 uvicorn[standard]
 torch
 torchvision
@@ -10,5 +10,4 @@ pytorch-msssim
 opencv-python
 tqdm
 torchsummary
-python-multipart
 requests

services/deblur.py

@@ -0,0 +1,58 @@
+# services/deblur.py
+import os
+import torch
+import torch.nn as nn
+import numpy as np
+from PIL import Image
+
+from models.fpn_inception import FPNInception
+
+# =====================
+# 初始化模型
+# =====================
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+print(f"🔹 [DeblurService] Using device: {device}")
+
+checkpoint_path = os.path.join("model", "deblurgan_v2_latest.pth")
+G = FPNInception(norm_layer=nn.InstanceNorm2d).to(device)
+
+checkpoint = torch.load(checkpoint_path, map_location=device)
+G.load_state_dict(checkpoint["G"], strict=False)
+G.eval()
+print(f"✅ [DeblurService] Model loaded from {checkpoint_path}")
+
+# ✅ 去模糊函式
+def deblur_image_tiled(img, tile_size=512, overlap=32):
+    w, h = img.size
+    new_w = (w // 32) * 32
+    new_h = (h // 32) * 32
+    if new_w != w or new_h != h:
+        img = img.resize((new_w, new_h), Image.BICUBIC)
+        w, h = new_w, new_h
+
+    img_np = np.array(img).astype(np.float32) / 255.0
+    img_tensor = torch.from_numpy(img_np).permute(2, 0, 1).unsqueeze(0).to(device)
+
+    stride = tile_size - overlap
+    tiles_x = list(range(0, w, stride))
+    tiles_y = list(range(0, h, stride))
+    if tiles_x[-1] + tile_size > w:
+        tiles_x[-1] = w - tile_size
+    if tiles_y[-1] + tile_size > h:
+        tiles_y[-1] = h - tile_size
+
+    output = torch.zeros_like(img_tensor)
+    weight = torch.zeros_like(img_tensor)
+
+    with torch.no_grad():
+        for y in tiles_y:
+            for x in tiles_x:
+                patch = img_tensor[:, :, y:y+tile_size, x:x+tile_size]
+                pred = G(patch)  # ✅ 改用 services 裡的模型 G
+                output[:, :, y:y+tile_size, x:x+tile_size] += pred
+                weight[:, :, y:y+tile_size, x:x+tile_size] += 1.0
+
+    output /= weight
+    output = torch.clamp(output, 0, 1)
+    out_np = (output.squeeze().permute(1, 2, 0).cpu().numpy() * 255.0).astype(np.uint8)
+    return Image.fromarray(out_np)