Update app.py

app.py

@@ -219,27 +219,33 @@ def process_image(input_image):
     """
     if input_image is None:
         return None, None
-    
+
     # Move model to GPU for processing
     MODEL.to('cuda')
     MODEL.eval()
-    
+
     # Convert from RGB to BGR for depth processing
     input_bgr = cv2.cvtColor(np.array(input_image), cv2.COLOR_RGB2BGR)
-    
+
     with torch.no_grad():
         # Get depth map
         depth = MODEL.infer_image(input_bgr)
-        
+
+        # **Apply Gaussian Blur to smooth the depth map**
+        kernel_size = (15, 15)  # Size of the Gaussian kernel (must be odd and positive)
+        sigma = 0  # If 0, sigma is calculated based on kernel size
+        depth = cv2.GaussianBlur(depth, kernel_size, sigma)
+        print(f"Applied Gaussian Blur with kernel size {kernel_size} and sigma {sigma}")
+
         # Normalize depth for visualization (0-255)
         depth_normalized = ((depth - depth.min()) / (depth.max() - depth.min()) * 255).astype(np.uint8)
-    
+
     # Move model back to CPU
     MODEL.to('cpu')
-    
+
     # Get intensity map
     intensity_map = get_image_intensity(np.array(input_image), gamma_correction=1.0)
-    
+
     # Blend depth raw with intensity map
     blended_result = blend_numpy_images(
         cv2.cvtColor(depth_normalized, cv2.COLOR_GRAY2RGB),  # Convert depth to RGB
@@ -247,10 +253,10 @@ def process_image(input_image):
         blend_factor=0.25,
         mode="normal"
     )
-    
+
     # Generate normal map from blended result
     normal_map = process_normal_map(blended_result)
-    
+
     return depth_normalized, normal_map
 
 @spaces.GPU