Update app.py

app.py

@@ -33,17 +33,29 @@ model = model.to(device)
 processor = DPTImageProcessor.from_pretrained("Intel/dpt-swinv2-tiny-256")
 
 def preprocess_image(image):
-    image = cv2.resize(image, (128, 128))
+    image = cv2.resize(image, (64, 36))  # 16:9 aspect ratio
     image = torch.from_numpy(image).permute(2, 0, 1).unsqueeze(0).float().to(device)
     return image / 255.0
 
-def plot_depth_map(depth_map, azimuth):
+def plot_depth_map(depth_map, original_image):
     fig = plt.figure(figsize=(16, 9))
     ax = fig.add_subplot(111, projection='3d')
     x, y = np.meshgrid(range(depth_map.shape[1]), range(depth_map.shape[0]))
-    ax.plot_surface(x, y, depth_map, cmap='viridis')
-    ax.view_init(elev=90, azim=azimuth)  # Look down onto the depth map
+    
+    # Flip x and y to mirror the plot
+    x = np.flip(x, axis=1)
+    y = np.flip(y, axis=0)
+    
+    # Rotate by 90 degrees
+    x, y = y, x
+    
+    # Use original image colors
+    colors = original_image.reshape(-1, 3) / 255.0
+    
+    ax.plot_surface(x, y, depth_map, facecolors=colors, shade=False)
+    ax.view_init(elev=70, azim=0)  # Tilt by 20 degrees (90 - 20 = 70)
     ax.set_zlim(0, 1)
+    plt.axis('off')
     plt.close(fig)
     
     fig.canvas.draw()
@@ -53,7 +65,7 @@ def plot_depth_map(depth_map, azimuth):
     return img
 
 @torch.inference_mode()
-def process_frame(image, azimuth):
+def process_frame(image):
     if image is None:
         return None
     preprocessed = preprocess_image(image)
@@ -62,11 +74,11 @@ def process_frame(image, azimuth):
     
     depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
     
-    return plot_depth_map(depth_map, azimuth)
+    return plot_depth_map(depth_map, cv2.resize(image, (64, 36)))
 
 interface = gr.Interface(
     fn=process_frame,
-    inputs=[gr.Image(sources="webcam", streaming=True), gr.Slider(0, 360, step=1)],
+    inputs=gr.Image(sources="webcam", streaming=True),
     outputs="image",
     live=True
 )