Reproduce “ScaleLSD: Scalable Deep Line Segment Detection Streamlined” 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
Zeran Ke, Bin Tan, Xianwei Zheng, Yujun Shen, Tianfu Wu, Nan Xue
ScaleLSD learns dense line geometry from unlabeled images and decodes it into sparse line segments, using classical LSD to correct direction estimates during pseudo-label generation. Trained on over 10 million images, it achieves zero-shot RDNIM structural Rep-5 of 0.337, above HAWPv3 and LSD, though HAWPv3 has lower localization error. On seven DTU scenes, its ACC-L is 0.7091, below LSD’s 0.8479.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.