Reproduce “Training-Free Acceleration of Multi-View 3D Transformers via Regime-Aware Attention Approximation” 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
Muhammed Izzet Sezer, Sevval Belkis Dikkaya
RAWATTA accelerates multi-view 3D reconstruction from unposed images by matching attention computation and precision to layer roles. It uses low-precision attention and, in RAWATTA-Fast, frame-attention and patch-merging operators while preserving dense matching and first-frame keys. On an RTX 4070 SUPER, RAWATTA is 2.08× faster at 50 images, with a −1.03 AUC@3 worst-dataset change at 10 frames; schedules are not universal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.