SOTAVerified

Deep Global Registration

2020-04-24CVPR 2020Code Available1· sign in to hype

Christopher Choy, Wei Dong, Vladlen Koltun

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

We present Deep Global Registration, a differentiable framework for pairwise registration of real-world 3D scans. Deep global registration is based on three modules: a 6-dimensional convolutional network for correspondence confidence prediction, a differentiable Weighted Procrustes algorithm for closed-form pose estimation, and a robust gradient-based SE(3) optimizer for pose refinement. Experiments demonstrate that our approach outperforms state-of-the-art methods, both learning-based and classical, on real-world data.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
3DLoMatch (10-30% overlap)DGR (reported in REGTR)Recall ( correspondence RMSE below 0.2)48.7Unverified
3DMatch (at least 30% overlapped - FCGF setting)DGR (RE (all), TE(all) are reported in PCAM)Recall (0.3m, 15 degrees)91.3Unverified
3DMatch (at least 30% overlapped - sample 5k interest points)DGR (reported in REGTR)Recall ( correspondence RMSE below 0.2)85.3Unverified
KITTI (FCGF setting)DGR + ICP (RE (all), TE(all) are reported in PCAM)Recall (0.6m, 5 degrees)98.2Unverified
KITTI (FCGF setting)DGR (RE (all), TE(all) are reported in PCAM)Recall (0.6m, 5 degrees)96.9Unverified

Reproductions