Reproduce “Toward Multimodal Image-to-Image Translation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A. Efros, Oliver Wang, Eli Shechtman

BicycleGAN generates multiple outputs for an input image in paired image-to-image translation. It combines a path that encodes a real target and reconstructs it with one that generates from a sampled latent code and recovers that code. On Google maps-to-satellites, it reports a 34.33 ± 2.69% AMT fooling rate and 0.110 ± 0.002 LPIPS, versus 27.93 ± 2.40% and 0.013 ± 0.000 for pix2pix+noise, showing higher measured realism and diversity on this task.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.