/semit

SEMi-supervised Image-to-image Translation

Primary LanguageJupyter NotebookMIT LicenseMIT

GT CS7641 SP2020 Project - Team 3 Learning Machines
Domain Adaptation via Semi-supervised Image-to-image Translation

Project GitHub Pages

https://ast0414.github.io/semit/

Presentation Video

Presentation

Abstract (of abstract)

In this project, we propose a (semi-)supervised image-to-image translation framework as a domain adaptation method based on a combination of variational autoencoders and generative adversarial networks upon a shared latent space assumption.

Codes

Results Preview

Reconstruction and Translation

Dataset Input Reconstruction Translation
MNIST
K-MNIST

t-SNE Visualization of the Shared Latent Space

tsne_latent