[Feature] Non-Random Slices for Locational Encoding of Input Image about aphantasia HOT 1 CLOSED

i used exactly that trick in another colab - CLIP+SIREN, sync_cut option.
if you look at the code of this repo, you'll notice it also has some legacy of that:

• slice_imgs takes list of images as input, to process them in such sync,
• sliced result is addressed by index [-1] (input was addressed as [0]), etc.

alas, it didn't work well enough with FFT, therefore SSIM loss has been used instead.

from aphantasia.