Skip to content

[Question] Finding interesting directions in the latent space #12

Open
@jpiabrantes

Description

@jpiabrantes

I was expecting that each dimension in the latent space (of the VAE) would represent some structure of the font e.g. boldness, italicness, etc etc.

However, I am seeing very little change when changing only one direction. I am zeroing all the dimensions, and then going from -3 to 3 on one dimension. I chose 3, because the KL loss should make each dimension dist a gaussian with mu=0 and sigma=1 and 3sigma should give us a good range to see what that dim is about.

Any idea why I am seeing very little changes in the image by doing this? Any other idea on how to automatically find interesting dimensions in the latent space?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions