Update app.py

app.py

@@ -11,6 +11,7 @@ from utils.dc_utils import read_video_frames, save_video
 from huggingface_hub import hf_hub_download
 
 # Examples for the Gradio Demo
+# Hier wurden die zusätzlichen Parameter (stitch, grayscale, blur) mit Standardwerten ergänzt.
 examples = [
     ['assets/example_videos/davis_rollercoaster.mp4', -1, -1, 1280, False, False, 0],
     ['assets/example_videos/Tokyo-Walk_rgb.mp4', -1, -1, 1280, False, False, 0],
@@ -62,7 +63,7 @@ def infer_video_depth(
     stitch: bool = False,
     grayscale: bool = False,
     blur: float = 0.0,
-    *,  # Ab hier folgen keyword-only Parameter:
+    *,  # Keyword-only parameters folgen ab hier:
     output_dir: str = './outputs',
     input_size: int = 518,
 ):
@@ -96,11 +97,10 @@ def infer_video_depth(
                 depth_vis = np.stack([depth_norm] * 3, axis=-1)
             else:
                 cmap = cm.get_cmap("inferno")
-                # cmap returns RGBA; we use only the first 3 channels and scale to 255
                 depth_vis = (cmap(depth_norm / 255.0)[..., :3] * 255).astype(np.uint8)
             # Apply Gaussian blur if requested (blur factor > 0)
             if blur > 0:
-                kernel_size = int(blur * 20) * 2 + 1  # ensures odd kernel size
+                kernel_size = int(blur * 20) * 2 + 1  # ensures an odd kernel size
                 depth_vis = cv2.GaussianBlur(depth_vis, (kernel_size, kernel_size), 0)
             # Concatenate side-by-side: RGB frame on the left, processed depth on the right
             stitched = cv2.hconcat([rgb_frame, depth_vis])
@@ -123,8 +123,8 @@ def construct_demo():
         
         with gr.Row(equal_height=True):
             with gr.Column(scale=1):
-                # Verwende hier source="upload" für das Datei-Upload-Feld
-                input_video = gr.Video(label="Input Video", source="upload")
+                # Hier verwenden wir den Video-Component ohne den 'source'-Parameter.
+                input_video = gr.Video(label="Input Video")
             with gr.Column(scale=2):
                 with gr.Row(equal_height=True):
                     processed_video = gr.Video(label="Preprocessed Video", interactive=False, autoplay=True, loop=True, show_share_button=True, scale=5)
@@ -149,7 +149,8 @@ def construct_demo():
             inputs=[input_video, max_len, target_fps, max_res, stitch_option, grayscale_option, blur_slider],
             outputs=[processed_video, depth_vis_video, stitched_video],
             fn=infer_video_depth,
-            cache_examples="lazy",
+            cache_examples=True,
+            cache_mode="lazy",
         )
         
         generate_btn.click(