Reproduce “Active Token Mixer” 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

Guoqiang Wei, Zhizheng Zhang, Cuiling Lan, Yan Lu, Zhibo Chen

Active Token Mixer (ATM) lets each channel select context locations along horizontal and vertical axes, then adaptively fuse those views with the feature. Its computation is O(HWC²), linear in spatial resolution and independent of receptive-field size. On ADE20K validation with Semantic FPN, ATMNet-L reaches 48.1 mIoU versus Twins-L’s 46.7, with 79.8M versus 103.7M parameters and 86.6G versus 102G FLOPs.

Summary by alphaXiv.

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