Reproduce “Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimation” 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

He Wang, Srinath Sridhar, Jingwei Huang, Julien Valentin, Shuran Song, Leonidas J. Guibas

Researchers at Stanford University and collaborators developed a method to estimate 6D pose and size for unseen object instances within a category using RGB-D images. The approach introduces Normalized Object Coordinate Space (NOCS) and a context-aware mixed-reality dataset, achieving 10.2% mAP for the (5°, 5 cm) 6D pose metric on real-world data.

Key insights

  • The Normalized Object Coordinate Space (NOCS) serves as an effective canonical representation for generalizing 6D pose and size estimation to diverse, unseen instances within object categories.
  • Context-aware mixed-reality data generation, which embeds synthetic objects realistically into real scenes, is crucial for bridging the domain gap and training robust models for real-world performance.
  • Predicting NOCS coordinates via pixel classification rather than direct regression, along with incorporating a specific symmetry loss, significantly improves pose estimation accuracy, especially for symmetric objects.

Summary by alphaXiv.

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