Research map / 3D vision / Depth and stereo
ASLFeat: Learning Local Features of Accurate Shape and Localization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 peakiness-based detection scoring (ablation) | D2-Net · HPatches | MMA@3: 45.36 -> 48.54 | Helped |
| Restore resolution and fuse detections across feature levels with MulDet (ablation) | ASLFeat · HPatches | MMA@3: 48.54 -> 68.66 | Helped |
| Use image-pyramid multi-scale detection instead of MulDet (ablation) | ASLFeat without DCN · HPatches | MMA@3: 48.54 -> 48.72 | No clear effect |
| Use in-network multi-scale detection instead of MulDet (ablation) | ASLFeat without DCN · HPatches | MMA@3: 48.54 -> 47.94 | No clear effect |
| Use MulDet instead of U-Net feature fusion (ablation) | ASLFeat without DCN · HPatches | MMA@3: 66.42 -> 68.66 | Helped |
| Use free-form DCN rather than similarity-constrained DCN (ablation) | ASLFeat with MulDet · HPatches | MMA@3: 71.67 -> 72.26 | Helped |
| Use free-form DCN rather than affine-constrained DCN (ablation) | ASLFeat with MulDet · HPatches | MMA@3: 71.80 -> 72.26 | Helped |
| Use free-form DCN rather than homography-constrained DCN (ablation) | ASLFeat with MulDet · HPatches | MMA@3: 71.89 -> 72.26 | Helped |
| Stack multiple DCN layers instead of using one (ablation) | ASLFeat with MulDet · HPatches | MMA@3: 71.08 -> 72.26 | Helped |
| Add explicit image-pyramid multi-scale inference (ablation) | ASLFeat · HPatches | MMA@3: 72.26 -> 72.64 | Helped |
| Train end-to-end with DCN instead of using two-stage training (ablation) | ASLFeat (MS) · HPatches | MMA@3: 72.64 -> 70.45 | Hurt |
| Add OANet learned matching | ASLFeat · Aachen Day-Night | Localization success (0.5m, 2°): 45.9 -> 48.0 | Helped |
Closest papers by meaning
- Few-Shot Geometry-Aware Keypoint Localization (CVPR 2023 · similarity 0.87)
- Enhancing Deformable Local Features by Jointly Learning To Detect and Describe Keypoints (CVPR 2023 · similarity 0.87)
- R2D2: Reliable and Repeatable Detector and Descriptor (NeurIPS 2019 · similarity 0.86)
- Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Features (CVPR 2024 · similarity 0.86)
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · similarity 0.85)
- RDD: Robust Feature Detector and Descriptor using Deformable Transformer (CVPR 2025 · 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.