Deep Generative Models that I have implemented, trained and experimented with.
Each directory is an independent project and repository. This repo is just a hub for them, but there are small descriptions and samples:
Pix2Pix
The task is to translate picture to a different modality (e.g. day photo to night one) using conditional GAN. Results:
StarGAN
This repository is an implementation of model described in StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. Results:
Experimenting w/ GLOW model
The main experiment is described in Why Normalizing Flows Fail to Detect Out-of-Distribution Data. For OOD SVHN dataset is used.
The graph shows that flows are biased towards learning graphical properties of the data such as local pixel correlations (e.g. nearby pixels usually have similar colors) rather than semantic properties of the data (e.g. what objects are shown in the image).
Quality results can be found in the GLOW repository.
VAE
Original VAE implementation. Results: