Research map / Architectures and efficiency / RNNs and normalization
EdgePoint2: Compact Descriptors for Superior Efficiency and Accuracy
Haodi Yao, Fenghua He, Ning Hao, Xie Chen
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 |
|---|---|---|---|
| replace PCA with LRA compression for L_op distillation (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 50.89 -> 51.75 | Improved |
| use Orthogonal Procrustes loss with LRA instead of HF-Net distillation with LRA (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 48.13 -> 51.75 | Improved |
| use Orthogonal Procrustes loss instead of Similarity-Preserving distillation (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 44.53 -> 51.75 | Improved |
| use Orthogonal Procrustes loss with LRA instead of AWDesc distillation (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 51.18 -> 51.75 | Improved |
| add image augmentation (N=4) to the Procrustes loss (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 51.75 -> 52.17 | Improved |
| add similarity loss L_sim (ablation) | EdgePoint2-M48 · MegaDepth-1500 ablation (EdgePoint2-M48) | AUC@5: 52.17 -> 52.07 | Mixed |
| increase model size from T32 to M32 | EdgePoint2 · MegaDepth-1500 | AUC@5: 46.12 -> 49.75 | Improved |
| increase descriptor dimension from 32 to 48 (M) | EdgePoint2 · MegaDepth-1500 | AUC@5: 49.75 -> 52.07 | Improved |
| replace EdgePoint with EdgePoint2-T32 (32-dim) | EdgePoint · MegaDepth-1500 | AUC@5: 42.92 -> 46.12 | Improved |
| replace EdgePoint with EdgePoint2-T32 | EdgePoint · MegaDepth-1500 | FPS on Orin-NX GPU: 838.8 -> 508.7 | Hurt |
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.
- XFeat — XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024) · better on 5 of 5 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
- P2-Net: Joint Description and Detection of Local Features for Pixel and Point Matching (ICCV 2021 · similarity 0.88)
- Deep Corner (IJCV 2023 · similarity 0.88)
- R2D2: Reliable and Repeatable Detector and Descriptor (NeurIPS 2019 · similarity 0.88)
- DeepPointMap: Advancing LiDAR SLAM with Unified Neural Descriptors (AAAI 2024 · similarity 0.86)
- Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Features (CVPR 2024 · similarity 0.85)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.85)
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.
