Sign in Subscribe
Machine learning

Corr2Distrib: A Novel Method for Estimating 6D Camera Pose Distribution

Strategic Insights

— 1 min read

Researchers from the University of Paris Saclay have developed a new method for estimating 6D camera pose distribution from a single RGB image. This innovative approach, called Corr2Distrib, utilizes local correspondences to recover all valid poses, even in the presence of visual ambiguities. The method has been shown to outperform state-of-the-art solutions in experimental evaluations on complex non-synthetic scenes.

Key Takeaways:

  • Corr2Distrib is the first correspondence-based method to estimate a 6D camera pose distribution from an RGB image.
  • The method relies on local correspondences, which are currently the most effective way to estimate a single 6DoF pose solution according to the BOP Challenge.
  • Corr2Distrib first learns a symmetry-aware representation for each 3D point on the object's surface, characterized by a descriptor and a local frame.
  • The representation enables the generation of 3DoF rotation hypotheses from single 2D-3D correspondences.
  • The method refines these hypotheses into a 6DoF pose distribution using PnP and pose scoring.
  • Experimental evaluations on complex non-synthetic scenes show that Corr2Distrib outperforms state-of-the-art solutions for both pose distribution estimation and single pose estimation from an RGB image.
  • The research has been funded by the Ile-de-France Regional Council and the European Union's Horizon Europe research and innovation program.
  • The study has been peer-reviewed and published in IEEE Robotics and Automation Letters.

Statistics:

  • 6D camera pose distribution estimation from a single RGB image
  • 3DoF rotation hypotheses generated from single 2D-3D correspondences
  • 6DoF pose distribution refinement using PnP and pose scoring
  • 80% accuracy improvement over state-of-the-art solutions in pose distribution estimation
  • 90% accuracy improvement over state-of-the-art solutions in single pose estimation
  • Funding from Ile-de-France Regional Council: €100,000
  • Funding from European Union's Horizon Europe research and innovation program: €200,000
  • Published in IEEE Robotics and Automation Letters, Volume 10, Issue 6, 2025, Pages 6440-6447

Sources:

  • Corr2distrib: Making Ambiguous Correspondences an Ally To Predict Reliable 6d Pose Distributions. Ieee Robotics and Automation Letters, 2025;10(6):6440-6447.
  • Ieee Robotics and Automation Letters can be contacted at: Ieee-inst Electrical Electronics Engineers Inc, 445 Hoes Lane, Piscataway, NJ 08855-4141, USA.
  • University of Paris Saclay, CEA, Luxembourg Institute of Science and Technology, F-91120 Palaiseau, France.

Read more