Research map / Architectures and efficiency / CNN design
FeatureBooster: Boosting Feature Descriptors With a Lightweight Neural Network
Read the paperCheck 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 |
|---|---|---|---|
| Boost ORB descriptors with FeatureBooster | ORB · Aachen Day-Night V1.1 Night | localized queries at the reported thresholds: 31.9 / 37.2 / 49.2 -> 49.2 / 61.8 / 73.3 | Helped |
| Boost SIFT descriptors with FeatureBooster | SIFT · InLoc DUC1 | localized queries at the reported thresholds: 29.3 / 43.4 / 51.5 -> 32.8 / 47.5 / 57.6 | Helped |
| Boost SuperPoint descriptors with FeatureBooster | SuperPoint · InLoc DUC2 | localized queries at the reported thresholds: 38.2 / 55.0 / 65.6 -> 40.5 / 58.0 / 67.9 | Helped |
| Add descriptor MLP self-boosting (ablation) | SuperPoint FeatureBooster · HPatches | MMA @3 / @5: 0.654 / 0.738 -> 0.654 / 0.738 | No clear effect |
| Add geometric encoding to descriptor MLP self-boosting (ablation) | SuperPoint FeatureBooster · HPatches | MMA @3 / @5: 0.654 / 0.738 -> 0.655 / 0.739 | Helped |
| Add Transformer cross-boosting (ablation) | SuperPoint FeatureBooster · HPatches | MMA @3 / @5: 0.654 / 0.738 -> 0.657 / 0.742 | Helped |
| Use both geometric encoding and Transformer cross-boosting (ablation) | SuperPoint FeatureBooster · HPatches | MMA @3 / @5: 0.654 / 0.738 -> 0.669 / 0.758 | Helped |
| Replace vanilla multi-head attention with AFT-Simple (ablation) | ORB+Boost-B · HPatches | MMA @3 / @5: 0.437 / 0.500 -> 0.436 / 0.495 | Hurt |
| Replace vanilla multi-head attention with AFT-Simple (ablation) | SuperPoint+Boost-F · HPatches | MMA @3 / @5: 0.679 / 0.777 -> 0.669 / 0.758 | Hurt |
| Replace vanilla multi-head attention with AFT-Simple (ablation) | FeatureBooster · RTX 3090, 2000 features | runtime: 4.9 -> 3.2 | Helped |
| Remove L_BOOST from training (ablation) | SIFT FeatureBooster · Rotated HPatches | MMA @3 / @5: 0.391 / 0.428 -> 0.216 / 0.236 | Hurt |
| Set L_BOOST weight to 1 instead of 10 (ablation) | SIFT FeatureBooster · Rotated HPatches | MMA @3 / @5: 0.391 / 0.428 -> 0.263 / 0.287 | Hurt |
Closest papers by meaning
- Local Descriptors Optimized for Average Precision (CVPR 2018 · similarity 0.87)
- Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task (CVPR 2020 · similarity 0.87)
- Comparative Evaluation of Hand-Crafted and Learned Local Features (CVPR 2017 · similarity 0.85)
- OmniGlue: Generalizable Feature Matching with Foundation Model Guidance (CVPR 2024 · similarity 0.85)
- HPatches: A Benchmark and Evaluation of Handcrafted and Learned Local Descriptors (CVPR 2017 · similarity 0.85)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.83)
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.