Reproduce “Dual-Manifold Variational Autoencoders” 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

Hongxu Chen, Chun Wang, Yanghao Wang, Hongxiang Li, Zhen Wang, Zhixiang Zhang, Long Chen

DM-VAE develops a latent diffusion tokenizer that retains reconstruction detail while making representations easier to generate. It couples a high-capacity latent branch to a lower-capacity structural branch, which trains generator features and guides structure at inference while leaving the detail residual unchanged. With DiT-XL after 400 epochs, it reports rFID 0.12 and FID50K 1.28 on ImageNet-1K; evidence is limited to tested settings.

Summary by alphaXiv.

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