debug

app.py

@@ -104,6 +104,9 @@ def vggsfm_demo(
     glbfile = target_dir + "/glbscene.glb"
     glbscene.export(file_obj=glbfile)    
 
+    del predictions
+    gc.collect()
+    torch.cuda.empty_cache()
     
     print(input_image)
     print(input_video)
@@ -199,7 +202,7 @@ with gr.Blocks() as demo:
         <li>upload the images (.jpg, .png, etc.), or </li> 
         <li>upload a video (.mp4, .mov, etc.) </li> 
     </ul> 
-    <p>The reconstruction should take <strong> up to 1 minute </strong>. If both images and videos are uploaded, the demo will only reconstruct the uploaded images. By default, we extract <strong> 1 image frame per second from the input video </strong>. To prevent crashes on the Hugging Face space, we currently limit reconstruction to the first 20 image frames. </p> 
+    <p>The reconstruction should normally take <strong> up to 90 second </strong>. If both images and videos are uploaded, the demo will only reconstruct the uploaded images. By default, we extract <strong> 1 image frame per second from the input video </strong>. To prevent crashes on the Hugging Face space, we currently limit reconstruction to the first 20 image frames. </p> 
     <p>SfM methods are designed for <strong> rigid/static reconstruction </strong>. When dealing with dynamic/moving inputs, these methods may still work by focusing on the rigid parts of the scene. However, to ensure high-quality results, it is better to minimize the presence of moving objects in the input data. </p> 
     <p>If you meet any problem, feel free to create an issue in our <a href="https://github.com/facebookresearch/vggsfm" target="_blank">GitHub Repo</a> ⭐</p> 
     <p>(Please note that running reconstruction on Hugging Face space is slower than on a local machine.) </p>