fix2

handler.py

@@ -15,24 +15,28 @@ logger = logging.getLogger(__name__)
 
 class EndpointHandler:
     def __init__(self, path="."):
-        logger.info("🚀 [INIT] Starting GFPGAN + Real-ESRGAN handler...")
+        logger.info("🚀 [INIT] GFPGAN + Real-ESRGAN handler starting...")
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
         self.half = self.device == "cuda"
         self.path = path
 
-        # Model download URLs
-        self.gfpgan_model_url = "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth"
-        self.realesr_model_url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth"
+        # Model URLs
+        self.gfpgan_model_url = (
+            "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth"
+        )
+        self.realesr_model_url = (
+            "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth"
+        )
 
         # Local paths
         self.gfpgan_model_path = os.path.join(path, "GFPGANv1.4.pth")
         self.realesr_model_path = os.path.join(path, "realesr-general-x4v3.pth")
 
-        # Lazy init placeholders
+        # Lazy init
         self.bg_upsampler = None
         self.restorer = None
 
-        # Ensure models exist
+        # Ensure model files exist
         self._ensure_model(self.gfpgan_model_url, self.gfpgan_model_path)
         self._ensure_model(self.realesr_model_url, self.realesr_model_path)
 
@@ -48,11 +52,12 @@ class EndpointHandler:
             logger.info(f"✅ Saved to {local_path}")
 
     def _init_models(self):
-        """Lazy-load GFPGAN and RealESRGAN models only when needed."""
+        """Lazy-load GFPGAN and Real-ESRGAN models."""
         if self.bg_upsampler is None:
-            logger.info("🧩 Initializing Real-ESRGAN background upsampler...")
+            logger.info("🧩 Initializing Real-ESRGAN upsampler...")
             model = SRVGGNetCompact(
-                num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu'
+                num_in_ch=3, num_out_ch=3, num_feat=64,
+                num_conv=32, upscale=4, act_type="prelu"
             )
             self.bg_upsampler = RealESRGANer(
                 scale=4,
@@ -77,14 +82,13 @@ class EndpointHandler:
             logger.info("✅ Models ready!")
 
     def __call__(self, data):
-        """Run restoration on an input image (bytes or URL)."""
+        """Restore a face photo."""
         self._init_models()
 
-        image = None
         if isinstance(data, dict) and "inputs" in data:
             data = data["inputs"]
 
-        # Accept bytes, PIL, URL, or base64 (Gradio style)
+        # Load image
         if isinstance(data, (bytes, bytearray)):
             image = Image.open(io.BytesIO(data)).convert("RGB")
         elif isinstance(data, str) and data.startswith("http"):
@@ -96,14 +100,11 @@ class EndpointHandler:
             raise ValueError("Unsupported input type")
 
         input_img = np.array(image, dtype=np.uint8)
+
         cropped_faces, restored_faces, restored_img = self.restorer.enhance(
-            input_img,
-            has_aligned=False,
-            only_center_face=False,
-            paste_back=True
+            input_img, has_aligned=False, only_center_face=False, paste_back=True
         )
 
-        # Convert restored image to bytes for output
         _, buffer = cv2.imencode(".jpg", restored_img)
         output_bytes = io.BytesIO(buffer.tobytes())
 

requirements.txt

@@ -1,5 +1,5 @@
-torch>=2.1
-torchvision>=0.16
+torch==2.1.0
+torchvision==0.16.0
 gfpgan==1.3.8
 realesrgan==0.3.0
 basicsr==1.4.2