Reproduce “UDT: Reconciling U-Nets and Diffusion Transformers with Data-Adaptive Token Reduction” 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

Junno Yun, Yaşar Utku Alçalar, Mehmet Akçakaya

UDT (U-Net Diffusion Transformer) reduces computation in image diffusion transformers by merging feature-similar tokens during encoding, preserving feature width, and restoring token layout for full-resolution denoising. On ImageNet 256×256 without classifier-free guidance, UDT-XL/2 reaches FID 7.7 after 80 epochs, compared with SiT-XL/2’s 7.9 after 1,400 epochs; this compares epochs, not elapsed training time.

Summary by alphaXiv.

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