Research map / Detection and segmentation / CNN features
Learning and Aggregating Deep Local Descriptors for Instance-level Recognition
Giorgos Tolias, Tomas Jenicek, 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 |
|---|---|---|---|
| use fixed attention instead of learned attention (ablation) | HOW ResNet18 · Validation (mini GLD2) | mAP: 83.7 -> 84.8 | Improved |
| use contrastive metric-learning loss instead of cross entropy (ablation) | HOW ResNet18 · Validation (mini GLD2) | mAP: 75.5 -> 83.7 | Improved |
| use contrastive loss instead of cross entropy in full architecture (ablation) | HOW ResNet18 · Revisited Oxford | mAP: 69.9 -> 74.8 | Improved |
| use contrastive loss instead of cross entropy in full architecture (ablation) | HOW ResNet18 · Tiny-GLD2 | muAP: 71.1 -> 81.3 | Improved |
| add local smoothing before ASMK aggregation (ablation) | HOW ResNet18 · Validation (mini GLD2) | number of descriptors per image: 344.1 -> 263.8 | Improved |
| use ResNet50 backbone instead of ResNet18 | HOW · Revisited Oxford/Paris (+R1M), mean over all | mAP: 54.8 -> 58.0 | Improved |
| increase number of local features n from 1000 to 2000 | HOW ResNet50- · Revisited Oxford/Paris (+R1M), mean over all | mAP: 58.0 -> 60.1 | Improved |
| use HOW local descriptors with ASMK instead of DELF with ASMK | HOW ResNet50- · Revisited Oxford/Paris (+R1M), mean over all | mAP: 53.0 -> 58.0 | Improved |
| use local descriptors with ASMK instead of best global descriptor GeM-AP | HOW ResNet50- · Revisited Oxford/Paris (+R1M), mean over all | mAP: 49.9 -> 60.1 | Improved |
| use HOW ResNet50- instead of GEM_w on GLD2 | HOW · GLD2 classification (CLS3) | muAP: 28.7 -> 36.5 | Improved |
Closest papers by meaning
- Local Descriptors Optimized for Average Precision (CVPR 2018 · similarity 0.89)
- Deep Image Retrieval: Learning Global Representations for Image Search (ECCV 2016 · similarity 0.89)
- Descriptor Distillation: A Teacher-Student-Regularized Framework for Learning Local Descriptors (IJCV 2024 · similarity 0.89)
- Local Features and Visual Words Emerge in Activations (CVPR 2019 · similarity 0.88)
- HPatches: A Benchmark and Evaluation of Handcrafted and Learned Local Descriptors (CVPR 2017 · similarity 0.88)
- Interferences in Match Kernels (TPAMI 2016 · similarity 0.87)
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.
