Reproduce “EfficientPose: An efficient, accurate and scalable end-to-end 6D multi object pose estimation approach” 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

Yannick Bukschat, Marcus Vetter

EfficientPose delivers state-of-the-art 6D multi-object pose estimation, achieving 97.35% ADD(-S) accuracy on the Linemod benchmark while maintaining real-time inference speeds exceeding 26 FPS for multiple objects. The method extends the EfficientDet architecture and introduces a novel 6D augmentation strategy, enabling a single-shot, end-to-end solution.

Key insights

  • Direct 6D pose estimation methods can match or surpass the accuracy of complex 2D+PnP approaches when an effective augmentation strategy is applied.
  • The novel 6D Augmentation, which correctly adjusts ground truth 6D poses during image transformations, is critical for achieving high accuracy on small datasets like Linemod.
  • Building upon efficient 2D object detection architectures allows for scalable, real-time, and end-to-end 6D multi-object pose estimation with minimal additional computational overhead.

Summary by alphaXiv.

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