Implemented as seen in Song et al.
-
A lightweight Unet model with time embeddings for use with both Probabalistic and Deterministic Diffusion Models. Able to obtain 95+ accuracy on FashionMNIST generation
-
Lightweight conditional VAE architecture with large latent space able to generalize well to FashionMNIST in just one epoch of training obtaining accuracies of 90+