Image Restoration

This marks the 10th month anniversary of the previous image restoration model.

The decoder model redraws the masked area based on the booru tags.

The input and the training target are based on:

• LAB color space
• FFT

As in another encoder model, the image was split to 4 equal-sized squares, and then one of the squares was masked by less than 30%.

The image goes through the PatchEmbed module. The transformer blocks receive the masked, patched image embeddings along with their Fourier transform, and the Camie tags as input.

Finally, the latent is upsampled to the original image size.

Datasets

• pixiv rank