Research map / Label-efficient and robust learning / Knowledge distillation
Descriptor Distillation: A Teacher-Student-Regularized Framework for Learning Local Descriptors
Yuzhen Liu, Qiulei Dong
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 |
|---|---|---|---|
| add teacher-student regularizer | HardNet · Brown | FPR@95%: 1.51 -> 1.23 | Improved |
| add teacher-student regularizer | SOSNet · Brown | FPR@95%: 1.03 -> 0.95 | Improved |
| add teacher-student regularizer | HyNet · Brown | FPR@95%: 0.84 -> 0.71 | Improved |
| add teacher-student regularizer (ablation) | DesDis-8 · Brown | FPR@95%: 3.66 -> 3.45 | Improved |
| add teacher-student regularizer (ablation) | DesDis-32 · Brown | FPR@95%: 1.64 -> 1.39 | Improved |
| add teacher-student regularizer (ablation) | DesDis-16 · Brown | FPR@95%: 2.20 -> 1.99 | Improved |
| use teacher distances instead of fixed distance supervision (ablation) | DesDis-HyNet · Brown (Liberty to YOS/ND) | FPR@95%: 0.73 -> 0.50 | Improved |
| add basic loss term L_B to the regularizer (ablation) | DesDis-HyNet · Brown (Liberty to YOS/ND) | FPR@95%: 0.64 -> 0.50 | Improved |
| increase light-weight student depth from 3 to 6 layers (ablation) | DesDis-32 family · Brown | FPR@95%: 3.03 -> 1.31 | Improved |
| add L_B when distilling SIFT (ablation) | DesDis-32 (SIFT teacher) · Brown | FPR@95%: 25.65 -> 1.67 | Improved |
| add TS regularizer with SIFT teacher (ablation) | DesDis-32 · Brown | FPR@95%: 1.64 -> 1.67 | No clear effect |
| distill ORB teacher into binary student (ablation) | Binary DesDis-32 · Brown | FPR@95%: 53.36 -> 56.06 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- HardNet — HARD-Net: Hardness-AwaRe Discrimination Network for 3D Early Activity Prediction (ECCV 2020) · better on 3 of 3 shared results
- SOSNet — SOSNet: Second Order Similarity Regularization for Local Descriptor Learning (CVPR 2019) · better on 1 of 1 shared result
- L2Net — L2-Net: Deep Learning of Discriminative Patch Descriptor in Euclidean Space (CVPR 2017) · better on 1 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
- Binary Online Learned Descriptors (TPAMI 2017 · similarity 0.89)
- Learning and Aggregating Deep Local Descriptors for Instance-level Recognition (ECCV 2020 · similarity 0.89)
- Revisiting Unsupervised Local Descriptor Learning (AAAI 2023 · similarity 0.88)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.88)
- NORM: Knowledge Distillation via N-to-One Representation Matching (ICLR 2023 · similarity 0.87)
- ScaleKD: Strong Vision Transformers Could Be Excellent Teachers (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.
