Reproduce “Efficient Fréchet Distributional Decoder Alignment for Latent Generative Models” 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
Xianghong Fang, Jiajun Zhu, Esma Aimeur, Dehan Kong, Tim G. J. Rudner
FDDA adapts the decoder in latent generation while keeping the generator fixed, training on generator-produced latents with a Fréchet distributional objective rather than reconstructions. On class-conditional ImageNet at 256 × 256, one epoch improves gFID and gFDr⁶ across 11 configurations; for TiTok-L-32, gFID drops from 2.75 to 1.34. The result is limited to these metrics and setting, and reconstruction quality declines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.