Update app.py

app.py

@@ -17,6 +17,7 @@ Please refer to our [paper](https://arxiv.org/abs/2501.12375), [project page](ht
 def stitch_rgbd_videos(
     processed_video: str,
     depth_vis_video: str,
+    audio_video: str,
     max_len: int = -1,
     target_fps: int = -1,
     max_res: int = 1280,
@@ -37,6 +38,7 @@ def stitch_rgbd_videos(
         full_frames, target_fps = read_video_frames(processed_video, max_len, target_fps, max_res=-1)
         depths, _ = read_video_frames(depth_vis_video, max_len, target_fps, max_res=-1)
         
+        
         # For each frame, create a visual depth image from the inferenced depths.
         d_min, d_max = depths.min(), depths.max()
         stitched_frames = []
@@ -120,7 +122,7 @@ def stitch_rgbd_videos(
             "ffmpeg",
             "-y",
             "-i", stitched_video_path,
-            "-i", processed_video,
+            "-i", audio_video,
             "-c:v", "copy",
             "-c:a", "aac",
             "-map", "0:v:0",
@@ -144,7 +146,9 @@ def construct_demo():
             with gr.Column(scale=1):
                 # Video input component for file upload.
                 processed_video = gr.Video(label="Input Video")
-                depth_vis_video = gr.Video(label="Generated Depth Video")                    
+                depth_vis_video = gr.Video(label="Depth Video") 
+            with gr.Column(scale=1):
+                audio_video = gr.Video(label="Audio Source")
             with gr.Column(scale=2):
                 with gr.Row(equal_height=True):
                     stitched_video = gr.Video(label="Stitched RGBD Video", interactive=False, autoplay=True, loop=True, show_share_button=True, scale=5)
@@ -154,7 +158,7 @@ def construct_demo():
                 with gr.Accordion("Advanced Settings", open=False):
                     max_len = gr.Slider(label="Max process length", minimum=-1, maximum=1000, value=-1, step=1)
                     target_fps = gr.Slider(label="Target FPS", minimum=-1, maximum=30, value=-1, step=1)
-                    max_res = gr.Slider(label="Max side resolution", minimum=480, maximum=1920, value=1280, step=1)
+                    max_res = gr.Slider(label="Max side resolution", minimum=480, maximum=1920, value=1920, step=1)
                     stitch_option = gr.Checkbox(label="Stitch RGB & Depth Videos", value=True)
                     grayscale_option = gr.Checkbox(label="Output Depth as Grayscale", value=True)
                     convert_from_color_option = gr.Checkbox(label="Convert Grayscale from Color", value=True)
@@ -165,7 +169,7 @@ def construct_demo():
         
         generate_btn.click(
             fn=stitch_rgbd_videos,
-            inputs=[processed_video, depth_vis_video, max_len, target_fps, max_res, stitch_option, grayscale_option, convert_from_color_option, blur_slider],
+            inputs=[processed_video, depth_vis_video, audio_video, max_len, target_fps, max_res, stitch_option, grayscale_option, convert_from_color_option, blur_slider],
             outputs=stitched_video,
         )
     
@@ -173,4 +177,4 @@ def construct_demo():
 
 if __name__ == "__main__":
     demo = construct_demo()
-    demo.queue(max_size=2).launch()
+    demo.queue(max_size=4).launch()