How to generate the pictures in Fig.6 in your paper?

[Nutingnon]

Thanks to your great work!

I would like to know how to generate pictures of latent-embedding like Fig.6 in this paper?

Thank you very much.

[saiboxx]

Hi @Nutingnon

I apologize for the late reply.
I was on a longer vacation and additionally left academia, so I check GitHub only sporadically these days around.

The latent-embedding images are the result of a lucky coincidence: The encoder of the VAE (whose architecture is from the LDM repo) encodes the images into a 3 x 64 x 64 shape. The 64s are the spatial dimension, where I show that the structure is kept and the 3 is the number of channels, which I just use as RGB. You could also say that this is some kind of false color image.

Does this resolve your issue?

Cheers,
Tobias