Diffusion Autoencoders: Toward a Meaningful and Decodable Representation | CVPR 2022
If you have any copyright issues on video, please send us an email at [email protected]
0:00 Introduction
0:07 Meaningful & Decodable Representation
0:30 Vanilla Autoencoders
0:59 Diffusion autoencoder DiffAE
2:05 What's in Semantic Code vs Noise Map?
2:18 Noise map captures stochastic details
2:31 Semantic Code is Linearly Discriminative!
3:12 Attribute Manipulation
3:37 Smooth Semantic Interpolation
3:49 Faster DPMs!
4:06 Unconditional sampling