Hi, thank you for releasing this great work!
In the repository, you visualize the PCA of non-masked visual tokens mapped to RGB values
Could you please explain how this visualization was created?
I’m particularly interested in:
Which features or layer outputs were used for PCA
How the PCA components were converted into RGB values
Whether any normalization or scaling was applied before visualization
Hi, thank you for releasing this great work!
In the repository, you visualize the PCA of non-masked visual tokens mapped to RGB values
Could you please explain how this visualization was created?
I’m particularly interested in:
Which features or layer outputs were used for PCA
How the PCA components were converted into RGB values
Whether any normalization or scaling was applied before visualization