Reproduce “Flow-of-Thought: A Framework for Visual Reasoning” 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
Mariia Baidachna, Nicolas Pugeault
Flow-of-Thought (FoT) classifies rotation pairs by generating visible intermediate states and comparing endpoint reconstructions for rotated and mirrored hypotheses. Its frozen flow is trained on rotations, rather than direct pixel paths. On locked tests, it reaches 100.0% accuracy on Tetris-like shapes and 99.0% on colored shapes. On BLINK Multi-view transfer, it scores 72.2% versus 63.9% for endpoint-only, but intervals overlap.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.