Decoder weights for reconstructing 16x16 patches instead of 8x8

openai / DALL-E

PyTorch package for the discrete VAE used for DALL·E.

Other

10.76k stars 1.94k forks source link

Decoder weights for reconstructing 16x16 patches instead of 8x8 #51

Open yatsurama opened 2 years ago

yatsurama commented 2 years ago

Hi,

thanks a lot for your great work!

I wanted to ask whether you have been experimenting with other patch dimensions than 8x8 such as 16x16 (i. e. splitting a 224x224 image into 14x14 visual tokens instead of 28x28). And if so, whether you could share the resulting weights.

Thank you

NiccoloNiccoli commented 5 months ago

Hi, did you find anything?