Research map / 3D vision / Depth and stereo
HyNet: Learning Local Descriptor with Hybrid Similarity Measure and Triplet Loss
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 |
|---|---|---|---|
| use HyNet instead of SOSNet | local descriptor matching · UBC phototour mean | FPR@95: 1.03 -> 0.84 | Helped |
| add L2 norm regularisation (ablation) | HyNet · HPatches matching average | MAP: 53.58 -> 53.97 | Helped |
| replace inner product with hybrid similarity (ablation) | HyNet · HPatches matching average | MAP: 53.19 -> 53.97; improves from s by 1.87 | Helped |
| replace L2 distance with hybrid similarity (ablation) | HyNet · HPatches matching average | MAP: 52.10 -> 53.97; improves from d by 0.78 | Helped |
| use FRN instead of Batch Normalisation (ablation) | HyNet · HPatches matching average | MAP: 52.04 -> 53.97 | Helped |
| use FRN instead of Instance Normalisation (ablation) | HyNet · HPatches matching average | MAP: 52.47 -> 53.97 | Helped |
| combine FRN with HardNet descriptor training (ablation) | HardNet · HPatches matching average | MAP: +1.33 | Helped |
| combine FRN with SOSNet descriptor training (ablation) | SOSNet · HPatches matching average | MAP: +1.10 | Helped |
| use the direct positive-inner-product/negative-distance combination (ablation) | HyNet triplet loss · UBC verification, Liberty-trained; Notredame and Yosemite test | FPR@95 | No clear effect |
| stack triplet losses with different similarity measures (ablation) | HyNet triplet loss · UBC verification, Liberty-trained; Notredame and Yosemite test | FPR@95 | Hurt |
| set hybrid-similarity balance parameter alpha to 2 (ablation) | HyNet · HPatches matching | MAP | Helped |
| add SOSR to HyNet (ablation) | HyNet · UBC phototour mean | FPR@95: 0.84 -> 0.82 | Helped |
Closest papers by meaning
- PUMP: Pyramidal and Uniqueness Matching Priors for Unsupervised Learning of Local Descriptors (CVPR 2022 · similarity 0.87)
- Learning Local Image Descriptors With Deep Siamese and Triplet Convolutional Networks by Minimising Global Loss Functions (CVPR 2016 · similarity 0.86)
- Learning Deep Descriptors With Scale-Aware Triplet Networks (CVPR 2018 · similarity 0.86)
- Local Descriptors Optimized for Average Precision (CVPR 2018 · similarity 0.85)
- Working hard to know your neighbor's margins: Local descriptor learning loss (NeurIPS 2017 · similarity 0.85)
- SDGMNet: Statistic-Based Dynamic Gradient Modulation for Local Descriptor Learning (AAAI 2024 · similarity 0.84)
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.