Research map / 3D vision / Human pose estimation
SRPose: Two-view Relative Pose Estimation with Sparse Keypoints
Rui Yin, Yulun Zhang, Zherong Pan, Jianjun Zhu, Cheng Wang, Biao Jia
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 |
|---|---|---|---|
| add keypoint-similarity prior guidance to cross-attention (ablation) | SRPose · Matterport | translation <=1m accuracy (%): 69.6 -> 74.2 | Improved |
| use cross-attention instead of only self-attention (ablation) | SRPose · Matterport | translation <=1m accuracy (%): 26.1 -> 74.2 | Improved |
| add position encoder (ablation) | SRPose · Matterport | translation <=1m accuracy (%): 61.0 -> 74.2 | Improved |
| add intrinsic calibration of keypoint coordinates (ablation) | SRPose · MegaDepth | pose AUC@5: 1.5 -> 16.6 | Improved |
| replace robust estimator with direct regression from sparse keypoints | matcher-based pose estimation · ScanNet-1500 | total time (ms): 269.7 -> 28.5 (vs LightGlue) | Improved |
| use SRPose instead of dense matcher DKM on ScanNet | SRPose · ScanNet-1500 | pose AUC@20: 68.3 -> 56.8 | Hurt |
| use object prompt bounding box instead of depth and segmentation matchers | SRPose · HO3D | ADD: 17.8 -> 36.4 | Improved |
| apply SRPose vs RoMa+DPT in relocalization precision | SRPose · Niantic | relocalization precision: 22.8 -> 16.9 precision; VCRE 45.6 -> 46.4 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- LightGlue — LightGlue: Local Feature Matching at Light Speed (ICCV 2023) · better on 1 of 2 shared results
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 1 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
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.88)
- SparsePose: Sparse-View Camera Pose Regression and Refinement (CVPR 2023 · similarity 0.87)
- ProbPose: A Probabilistic Approach to 2D Human Pose Estimation (CVPR 2025 · similarity 0.87)
- UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image (CVPR 2025 · similarity 0.86)
- From Correspondences to Pose: Non-minimal Certifiably Optimal Relative Pose without Disambiguation (CVPR 2024 · similarity 0.86)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (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.
