force depth map to same dims as image

app.py

@@ -111,20 +111,6 @@ def generate_3d_model(depth, image_path, focallength_px):
 
 @spaces.GPU(duration=20)
 def predict_depth(input_image):
-    """
-    Predict the depth map from the input image, generate visualizations and a 3D model.
-
-    Args:
-        input_image (str): Path to the input image file.
-
-    Returns:
-        tuple:
-            - str: Path to the depth map image.
-            - str: Focal length in pixels or an error message.
-            - str: Path to the raw depth data CSV file.
-            - str: Path to the generated 3D model file for viewing.
-            - str: Path to the downloadable 3D model file.
-    """
     temp_file = None
     try:
         # Resize the input image to a manageable size
@@ -150,18 +136,20 @@ def predict_depth(input_image):
         if depth.ndim != 2:
             depth = depth.squeeze()
 
-        # **Downsample depth map and image to improve processing speed**
-        downscale_factor = 2  # Factor by which to downscale (e.g., 2 reduces dimensions by half)
-        depth = depth[::downscale_factor, ::downscale_factor]
-        # Convert image tensor to CPU and NumPy for slicing
-        image_np = image.cpu().detach().numpy()[0].transpose(1, 2, 0)
-        image_ds = image_np[::downscale_factor, ::downscale_factor, :]
-        # Update focal length based on downscaling
-        focallength_px = focallength_px / downscale_factor
+        # Print debug information
+        print(f"Depth shape: {depth.shape}")
+        print(f"Image shape: {image.shape}")
+
+        # Ensure depth and image have the same dimensions
+        if depth.shape != image.shape[2:]:
+            # Resize depth to match image dimensions
+            depth = np.resize(depth, image.shape[2:])
 
-        # **Note:** The downscaled image is saved back to the temporary file for consistency
-        downscaled_image = Image.fromarray((image_ds * 255).astype(np.uint8))
-        downscaled_image.save(temp_file)
+        # No downsampling
+        downscale_factor = 1
+        
+        # Convert image tensor to CPU and NumPy
+        image_np = image.cpu().detach().numpy()[0].transpose(1, 2, 0)
 
         # No normalization of depth map as it is already in meters
         depth_min = np.min(depth)