Research map / Label-efficient and robust learning / Recognition and long-tail
Memory-Based Neighbourhood Embedding for Visual Recognition
Suichan Li, Dapeng Chen, Bin Liu, Nenghai Yu, Rui Zhao
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 memory update during training (ablation) | MNE · CUHK03 | rank-1: 72.9 -> 77.3 | Improved |
| add memory update during training (ablation) | MNE · CUHK03 | mAP: 74.6 -> 77.7 | Improved |
| add BCE supervision of attention (K=12) (ablation) | MNE · CUHK03 | mAP: 73.0 -> 77.7 | Improved |
| add BCE supervision of attention (K=8) (ablation) | MNE · CUHK03 | mAP: 76.0 -> 76.8 | Improved |
| add BCE supervision of attention (K=8) (ablation) | MNE · miniImageNet 1-shot | accuracy: 58.51 -> 59.92 | Improved |
| add BCE supervision of attention (K=12) (ablation) | MNE · miniImageNet 1-shot | accuracy: 59.32 -> 60.04 | Improved |
| add BCE supervision of attention (K=12) (ablation) | MNE · miniImageNet 5-shot | accuracy: 71.54 -> 71.62 | No clear effect |
| add BCE supervision of attention (K=8) (ablation) | MNE · miniImageNet 5-shot | accuracy: 71.40 -> 71.76 | Improved |
| enlarge memory by using higher gallery sampling ratio (ablation) | MNE · CUHK03 | mAP and rank-1 | Improved |
| replace mean/max pooling with attentive aggregation (ablation) | MNE · CUHK03 and miniImageNet | accuracy | Improved |
| increase neighbour number K with mean/max aggregation (ablation) | MNE · CUHK03 and miniImageNet | accuracy | Hurt |
| use too small or too large K (ablation) | MNE · CUHK03 and miniImageNet | accuracy | Hurt |
Closest papers by meaning
- Improving Generalization via Scalable Neighborhood Component Analysis (ECCV 2018 · similarity 0.89)
- Self-Supervised Representation Learning via Neighborhood-Relational Encoding (ICCV 2019 · similarity 0.86)
- Zero-Shot Visual Recognition via Bidirectional Latent Embedding (IJCV 2017 · similarity 0.85)
- Compound Memory Networks for Few-shot Video Classification (ECCV 2018 · similarity 0.85)
- Unsupervised Deep Learning by Neighbourhood Discovery (ICML 2019 · similarity 0.85)
- Memory Matching Networks for One-Shot Image Recognition (CVPR 2018 · 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.
