Research map / Point clouds / Registration and matching
DeepGMR: Learning Latent Gaussian Mixture Models for Registration
Wentao Yuan, Benjamin Eckart, Kihwan Kim, Varun Jampani, Dieter Fox, Jan Kautz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learned latent GMM correspondences (DeepGMR) instead of point-to-point matching (DCP) on unseen categories | DeepGMR vs DCP · ModelNet unseen | RMSE: 0.34 -> 0.01 | Improved |
| learned latent GMM correspondences instead of point-to-point matching (DCP) on real scans | DeepGMR vs DCP · ICL-NUIM | Recall@0.2: 0.16 -> 0.99 | Improved |
| add sensor-like noise to source and target independently | DCP · ModelNet clean to noisy | RMSE: 0.02 -> 0.08 | Hurt |
| add independent noise to source and target | DeepGMR · ModelNet clean to noisy | RMSE: 0.00 -> 0.01 | No clear effect |
| probabilistic latent matching instead of exhaustive search (TEASER++) for speed | DeepGMR vs TEASER++ · ModelNet40 runtime | time per instance: 13.3s -> 11ms | Improved |
| use more GMM components J beyond 16 (ablation) | DeepGMR · ModelNet noisy | RMSE CDF | No clear effect |
| add RRI rotation-invariant input features instead of raw xyz (ablation) | DeepGMR · ModelNet noisy | RMSE CDF | Improved |
| train with RMSE metric loss instead of simple MSE loss (ablation) | DeepGMR · ModelNet noisy | RMSE CDF | No clear effect |
| test with 4 times denser point clouds without finetuning | DeepGMR · ModelNet noisy | RMSE CDF | No clear effect |
| test with 4 times sparser point clouds without finetuning | DeepGMR · ModelNet noisy | RMSE CDF | Hurt |
| increase FGR input points to 4096 on clean data | FGR · ModelNet clean | RMSE: 0.19 -> 0.04 | Improved |
| increase FGR input points to 4096 on noisy data | FGR · ModelNet noisy | RMSE: 0.2 -> 0.22 | No clear effect |
Later papers that compared with this method
4 later accepted papers report DeepGMR as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- PointNetLK Revisited (CVPR 2021 · better on 6 of 8 shared results)
- Canonical Capsules: Self-Supervised Capsules in Canonical Pose (NeurIPS 2021 · better on 3 of 4 shared results)
- Distinctiveness Oriented Positional Equilibrium for Point Cloud Registration (ICCV 2021 · better on 2 of 2 shared results)
- Robust Point Cloud Registration Framework Based on Deep Graph Matching (CVPR 2021 · better on 0 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HGMR: Hierarchical Gaussian Mixtures for Adaptive 3D Registration (ECCV 2018 · similarity 0.93)
- PointGMM: A Neural GMM Network for Point Clouds (CVPR 2020 · similarity 0.91)
- GOGMA: Globally-Optimal Gaussian Mixture Alignment (CVPR 2016 · similarity 0.88)
- Deep Closest Point: Learning Representations for Point Cloud Registration (ICCV 2019 · similarity 0.88)
- GRICP: Granular-Ball Iterative Closest Point with Multikernel Correntropy for Point Cloud Fine Registration (AAAI 2025 · similarity 0.86)
- FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
