Research map / Architectures and efficiency / CNN design
Understanding and Improving Kernel Local Descriptors
Arun Mukundan, Giorgos Tolias, Andrei Bursuc, Hervé Jeǵou, Ondřej Chum
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 Cartesian parametrization to polar | polar descriptor · Photo Tourism | FPR95 (%): 22.42 -> 25.37 | Hurt |
| add supervised whitening W_S to polar+cartes (ablation) | polar + cartes · Photo Tourism | FPR95 (%): 25.37 -> 5.94 | Improved |
| add Cartesian parametrization to polar under supervised whitening (ablation) | polar + W_S · Photo Tourism | FPR95 (%): 7.06 -> 5.94 | Improved |
| replace PCA+SQRT with learned supervised whitening (ablation) | polar · Photo Tourism | FPR95 (%): 8.30 -> 7.06 | Improved |
| use unsupervised attenuated whitening instead of supervised | polar + cartes · Photo Tourism | FPR95 (%): 5.94 -> 6.79 | Hurt |
| use shrinkage whitening W_US instead of attenuated W_UA (ablation) | polar + cartes · Photo Tourism | FPR95 (%): 6.79 -> 7.22 | No clear effect |
| add unsupervised whitening W_US (learned on PT) to raw descriptor (ablation) | polar + cartes · HPatches retrieval | mAP (%): 45.23 -> 53.50 | Improved |
| use unsupervised instead of supervised whitening when learned on PT (ablation) | polar + cartes · HPatches retrieval | mAP (%): 49.66 -> 53.50 | Improved |
| learn supervised whitening on HP instead of PT (ablation) | polar + cartes + W_S · HPatches retrieval | mAP (%): 49.66 -> 61.79 | Improved |
| learn unsupervised W_UA on HP instead of PT (ablation) | polar + cartes + W_UA · HPatches retrieval | mAP (%): 52.78 -> 56.36 | Improved |
| learn unsupervised W_US on HP instead of PT (ablation) | polar + cartes + W_US · HPatches verification | mAP (%): 78.81 -> 80.70 | Improved |
| vary shrinkage parameter t or beta (ablation) | W_UA / W_US · Photo Tourism and HPatches | FPR95 (%) | No clear effect |
Closest papers by meaning
- Explicit Spatial Encoding for Deep Local Descriptors (CVPR 2019 · similarity 0.90)
- Beyond Cartesian Representations for Local Descriptors (ICCV 2019 · similarity 0.89)
- Local Descriptors Optimized for Average Precision (CVPR 2018 · similarity 0.87)
- Convolutional Patch Representations for Image Retrieval: An Unsupervised Approach (IJCV 2016 · similarity 0.86)
- Kernelized Subspace Pooling for Deep Local Descriptors (CVPR 2018 · similarity 0.85)
- Learning and Aggregating Deep Local Descriptors for Instance-level Recognition (ECCV 2020 · 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.
