Reproduce “Scale as a Modality: Multi-Scale Pixel-Space Diffusion Transformers” 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

Ruibo Chen, Kaishen Wang, Tianyi Xiong, Chenxi Liu, Yanshuo Chen, Heng Huang, Jian Du

MS-pDiT generates pixel-space images by representing one noisy RGB image at several patch scales, allowing streams to exchange information through compact joint attention. In ImageNet 256×256 80-epoch ablations, a 4×4 intermediate stream yields FID 2.31, versus 3.63 when similar capacity goes only to the coarse stream. This supports distributing capacity across scales, not a universal gain or lower total computation.

Summary by alphaXiv.

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