Video autoencoding

[soumickmj]

Hi @man805
Great work, congratulations.

I'm really interested in your work, but for a different task.
I'm trying to use the code to perform perform video auto-encoding.
I do not need the editing part, only able to encode my video in a lower dimensional space.

We are interested in obtaining the identity feature zid for the full video, while also obtaining the motion features for z(n)lnd of each frame.

What I'm seeing from the Vox dataset class, randomly frames are chosen to form a batch. This is confusing me to understand how the continous motion feature is encoded.

Could you please guide me to obtain the identiy feature from the whole video and the motion feature for each frame from a given video?

Thank you so much.

With Regards,
Soumick