Reproduce “Leveraging 3D Geometric Priors in 2D Rotation Symmetry Detection” 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

Ahyun Seo, Minsu Cho

This research introduces a method for 2D rotation symmetry detection that leverages 3D geometric priors by directly predicting 3D rotation centers and supporting vertices. A key innovation is a "vertex reconstruction module" that enforces 3D geometric consistency, ensuring accurate and structurally sound detection. The model achieved a mean AP of 30.6 for rotation vertex detection and an F1-score of 33.2 on the DENDI dataset, outperforming prior state-of-the-art methods.

Key insights

  • Directly predicting 3D rotation parameters and then reconstructing all vertices based on these 3D priors significantly improves 2D symmetry detection accuracy and geometric consistency, demonstrating the value of enforcing 3D structure.
  • Simply lifting 2D features to 3D and independently regressing 3D points without explicit geometric enforcement can degrade performance due to noise or misalignment from the projection process.
  • A set-based detection approach, combined with 3D-aware feature learning using camera queries, provides a robust and end-to-end framework for comprehensive rotation symmetry detection.

Summary by alphaXiv.

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