Research map / Label-efficient and robust learning / Recognition and long-tail
Revisiting Unsupervised Local Descriptor Learning
Wufan Wang, Lei Zhang, Hua Huang
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 |
|---|---|---|---|
| Combine rule-based training with clustering-based training (ablation) | Local descriptor learning on UBC Phototour · UBC Phototour (mean; 128-dim) | FPR95 (%): 7.94 -> 3.23 | Improved |
| Add DHS to select transformation-rule hyperparameters (ablation) | Rule-based local descriptor training on UBC Phototour · UBC Phototour (mean; 128-dim) | FPR95 (%): 7.94 -> 6.54 | Improved |
| Add clustering after DHS-based rule training (ablation) | HybridDesc on UBC Phototour · UBC Phototour (mean; 128-dim) | FPR95 (%): 6.54 -> 2.98 | Improved |
| Replace full-dataset clustering with ODC (ablation) | HybridDesc on UBC Phototour · UBC Phototour (mean; 128-dim) | FPR95 (%): 2.98 -> 2.93 | Improved |
| Use ODC to cluster only ambiguous samples (ablation) | Clustering-based training on UBC Phototour · UBC Phototour | Clustering time / model optimization time (%): 54.07 -> 25.33 | Improved |
| Use ρ values of 0.5 or 0.6 for selective clustering (ablation) | ODC in HybridDesc · UBC Phototour | FPR95 (%) | Improved |
| Increase the distance-ratio threshold ρ (ablation) | ODC in HybridDesc · UBC Phototour | FPR95 (%) | Hurt |
| Increase clustering frequency (ablation) | HybridDesc · UBC Phototour | FPR95 (%) | Improved |
| Set the DHS weighting factor λ above 0.01 (ablation) | HybridDesc · UBC Phototour | FPR95 (%) | Hurt |
| Use HybridDesc instead of BLCD | Binary local descriptor learning · HPatches patch verification | mAP (%): 83.86% -> 83.97% | Improved |
| Use HybridDesc instead of BLCD-r | Real-valued local descriptor learning · HPatches image matching | mAP (%): 46.91% -> 50.15% | Improved |
| Use HybridDesc instead of BLCD | Binary local descriptor learning · HPatches patch retrieval | mAP (%): 53.94% -> 55.14% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- HardNet — HARD-Net: Hardness-AwaRe Discrimination Network for 3D Early Activity Prediction (ECCV 2020) · better on 0 of 1 shared result
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
- Descriptor Distillation: A Teacher-Student-Regularized Framework for Learning Local Descriptors (IJCV 2024 · similarity 0.88)
- Local Descriptors Optimized for Average Precision (CVPR 2018 · similarity 0.86)
- Beyond Cartesian Representations for Local Descriptors (ICCV 2019 · similarity 0.86)
- Learning Local Descriptors With a CDF-Based Dynamic Soft Margin (ICCV 2019 · similarity 0.85)
- Invariant Information Clustering for Unsupervised Image Classification and Segmentation (ICCV 2019 · similarity 0.85)
- Debiased Novel Category Discovering and Localization (AAAI 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.
