erow/video_representations

This is a demo of the VideoMAE model to visualize the attention map, latent space, and reconstruction of a video.

Choose one of the following modes to visualize the video:

• —Reconstruction: Reconstruct the video by masking 90% of the patches and reconstructing the masked patches.
• —Attention: Visualize the average attention map of the last layer.
• —Latent: Visualize the PCA components of the latent space of the video.

You can choose the model and load the example video or upload your own video to visualize the video.