Reproduce “Sample-Aligned Decoders 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
Yu Lei, Yi Li, Hongcheng Gao, Yipeng Wei, Liang Li, Li Su, Qingming Huang
SAD adapts the decoder in latent image generators after generator training, keeping the generator fixed. It aligns decoded generated latents with unpaired real-image features and anchors decoder outputs on encoder latents to preserve reconstruction. On ImageNet 256×256, it lowers REPA-E FID-50K from 1.15 to 0.98; gains span four tested generators, not every metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.