Reproduce “Attribute2Image: Conditional Image Generation from Visual Attributes” 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

Xinchen Yan, Jimei Yang, Kihyuk Sohn, Honglak Lee

The authors introduce disCVAE, an attribute-conditioned image generator that separates foreground and background variation. Attributes and a foreground latent code generate an object and gate, while a separate latent code generates the background; the gate combines them. On LFW face reconstruction, disCVAE reports pixel-level MSE of 10.0 ± 0.1 versus 11.8 ± 0.1 for vanilla CVAE, but this does not establish improved realism or that layering alone caused the difference.

Summary by alphaXiv.

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