Latent Diffusion Model without Variational Autoencoder

This research introduces SVG, a novel latent diffusion framework that eliminates variational autoencoders by leveraging frozen DINO features for structured l...

Level: advanced

By Minglei Shi and 8 other authors

Category: education