Reproduce “LitePT: Lighter Yet Stronger Point Transformer” 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

Yuanwen Yue, Damien Robert, Jianyuan Wang, Sunghwan Hong, Jan Dirk Wegner, Christian Rupprecht, Konrad Schindler

LitePT assigns convolution to early, high-resolution stages and attention to later, lower-resolution stages, adding parameter-free PointROPE to supply spatial information from 3D coordinates. On nuScenes validation, LitePT-S reaches 82.2 mIoU versus PTv3’s 80.4, with 21.5 versus 35.1 ms latency in an ablation. On full ScanNet, however, it scores 76.5 versus PTv3’s 77.5.

Summary by alphaXiv.

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