Optimizing Hierarchical Image VAEs for Sample Quality

Eric Luhman, Troy Luhman

While hierarchical variational autoencoders (VAEs) have achieved great density estimation on image modeling tasks, samples from their prior tend to look less convincing than models…