Research map / 3D vision / Human pose estimation
Scalable and Generalizable Correspondence Pruning via Geometry-Consistent Pre-Training
Tangfei Liao, Xiaoqin Zhang, Tao Wang, Hao Ye, Min Li, Guobao Xiao, Mang Ye
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 local stream to global-only encoder (ablation) | CorrFormer · YFCC100M outdoor | Pose AUC@5: 34.64 -> 37.69 | Improved |
| add global stream to local-only encoder (ablation) | CorrFormer · YFCC100M outdoor | Pose AUC@5: 28.65 -> 37.69 | Improved |
| add consensus interaction between streams (vs local-to-global without interaction) (ablation) | CorrFormer · YFCC100M outdoor | Pose AUC@5: 36.00 -> 37.69 | Improved |
| use implicit weight-shared interaction instead of cross-attention (ablation) | CorrFormer · YFCC100M outdoor | Pose AUC@5: 36.92 -> 37.69 | Improved |
| use 60% random masking instead of 40% (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 38.84 -> 39.54 | Improved |
| use 60% random masking instead of 80% (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 38.25 -> 39.54 | Improved |
| use random masking instead of block masking (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 38.75 -> 39.54 | Improved |
| add keypoint distribution loss (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 39.07 -> 39.54 | Improved |
| use ordered positional prompts instead of shuffled (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 38.02 -> 39.54 | Improved |
| add mask embeddings to positional prompts (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 39.05 -> 39.54 | Improved |
| pre-train on SIFT+NN correspondences with 89.23% outliers instead of ground-truth inliers (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 39.54 -> 37.73 | Hurt |
| pre-train on DKM+RANSAC correspondences (6.10% outliers) instead of GT inliers (ablation) | CorrMAE · YFCC100M outdoor | Pose AUC@5: 39.54 -> 39.75 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BCLNet — BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024) · better on 4 of 4 shared results
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
- VSFormer: Visual-Spatial Fusion Transformer for Correspondence Pruning (AAAI 2024 · similarity 0.89)
- BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024 · similarity 0.88)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (ICLR 2024 · similarity 0.88)
- Unifying Correspondence Pose and NeRF for Generalized Pose-Free Novel View Synthesis (CVPR 2024 · similarity 0.87)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.87)
- PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting for Novel View Synthesis (ICML 2025 · similarity 0.87)
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.
