Research map / Label-efficient and robust learning / Recognition and long-tail
Interferences in Match Kernels
Naila Murray, Hervé Jeǵou, Florent Perronnin, Andrew Zisserman
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 |
|---|---|---|---|
| replace sum aggregation with democratic aggregation (ablation) | T-embedding + RN · Oxford5k | mAP: 63.0 -> 67.0 | Improved |
| replace sum aggregation with GMP (ablation) | T-embedding + RN · Oxford5k | mAP: 63.0 -> 70.0 | Improved |
| replace sum aggregation with GMP (ablation) | T-embedding + RN · Oxford105k | mAP: 55.0 -> 64.4 | Improved |
| replace sum aggregation with GMP (ablation) | T-embedding + RN · INRIA Holidays | mAP: 73.4 -> 76.5 | Improved |
| add rotation and normalisation (RN) (ablation) | T-embedding + sum · Oxford5k | mAP: 58.1 -> 63.0 | Improved |
| add rotation and normalisation (RN) (ablation) | T-embedding + GMP · Oxford5k | mAP: 63.4 -> 70.0 | Improved |
| add rotation and normalisation (RN) (ablation) | FV + sum · INRIA Holidays | mAP: 61.7 -> 68.7 | Improved |
| replace sum aggregation with GMP (ablation) | FV + RN · INRIA Holidays | mAP: 68.7 -> 69.4 | Mixed |
| replace sum aggregation with democratic aggregation (ablation) | FV · INRIA Holidays | mAP: 61.7 -> 62.1 | No clear effect |
| replace FV embedding with T-embedding (ablation) | sum aggregation · Oxford5k | mAP: 49.5 -> 58.1 | Improved |
| reduce T-embedding GMP+RN vector from 8,064 to 128 dimensions (ablation) | T-embedding + GMP + RN · Oxford5k | mAP: 70.0 -> 42.2 | Hurt |
| reduce T-embedding GMP+RN vector from 8,064 to 1,024 dimensions (ablation) | T-embedding + GMP + RN · Oxford5k | mAP: 70.0 -> 58.3 | Hurt |
Closest papers by meaning
- Learning Deep Match Kernels for Image-Set Classification (CVPR 2017 · similarity 0.88)
- Efficient Large-Scale Similarity Search Using Matrix Factorization (CVPR 2016 · similarity 0.88)
- Improving Large-Scale Image Retrieval Through Robust Aggregation of Local Descriptors (TPAMI 2016 · similarity 0.87)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.86)
- MagicLens: Self-Supervised Image Retrieval with Open-Ended Instructions (ICML 2024 · similarity 0.85)
- When Kernels Multiply, Clusters Unify: Fusing Embeddings with the Kronecker Product (NeurIPS 2025 · 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.
