search

dangeng / visual_anagrams

Code for the paper "Visual Anagrams: Generating Multi-View Optical Illusions with Diffusion Models"
MIT License
870 stars 81 forks source link

Dataset details #1

Closed haofanwang closed 11 months ago

haofanwang commented 11 months ago

Interesting work!

Is there any plan to release the training dataset and code?

dangeng commented 11 months ago

Hey, thanks!

We don't actually do any training. We just use pretrained off-the-shelf diffusion models (in particular DeepFloyd IF).

Actually, I might be wrong, but I don't think there's that many of these multi-view optical illusions in the wild. Maybe a few hundred, if that? You would be hard pressed to make a dataset large enough to train a model. Which I think makes this method quite interesting!