Reproduce “Pixel is All You Need” 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
Jie Chu, Haoting Shi, Fucheng Zhang, Jiawei Liu, Junwei Liao, Ying Wen, Muning Wen
The authors train a pixel-only model to reason by updating a canvas frame by frame. On Compute Then Paint, it reaches 98.1% test accuracy at 130M parameters, compared with 31.1% for direct answering trained for four times as many steps. Success requires correct arithmetic, region labels and marks, and painting with IoU of at least 0.8; results cover generated tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.