Reproduce “Latent Traversals in Generative Models as Potential Flows” 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
Yue Song, T. Anderson Keller, Nicu Sebe, Max Welling
The authors introduce a method for editing images from pretrained generators by learning time-varying latent potentials whose gradients update latent codes while keeping the generator fixed. A wave-equation-inspired constraint and auxiliary losses shape the paths and encourage distinguishable image changes. On SNGAN trained on AnimeFace, it reaches 65.89% VP accuracy versus 58.83% for WarpedSpace, measuring path identification, not semantic purity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.