Research map / Detection and segmentation / Open-set and few-shot detection
RepMet: Representative-Based Metric Learning for Classification and Few-Shot Object Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly train the backbone, embedding, and class representatives | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 1-shot, no episode fine-tuning | mAP (%): 41.3 -> 56.9 | Helped |
| jointly train the backbone, embedding, and class representatives | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 5-shot, no episode fine-tuning | mAP (%): 58.2 -> 68.8 | Helped |
| jointly train the embedding and detector rather than train the embedding offline | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 1-shot, no episode fine-tuning | mAP (%): 19.0 -> 56.9 | Helped |
| fine-tune the embedding and representatives on episode training images (ablation) | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 1-shot | mAP (%): 56.9 -> 59.2 | Helped |
| fine-tune the embedding and representatives on episode training images (ablation) | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 5-shot | mAP (%): 68.8 -> 73.9 | Helped |
| fine-tune the embedding and representatives on episode training images (ablation) | RepMet few-shot detector · ImageNet-LOC, 214 unseen animal classes, 10-shot | mAP (%): 71.5 -> 79.2 | Helped |
| use trained class representatives instead of episode-example embeddings (ablation) | RepMet detector · ImageNet-LOC, 100 seen animal classes, 5-shot | mAP (%): 79.4 -> 86.3 | Helped |
| replace the standard FPN-DCN classifier head with the RepMet DML head | object detection · PASCAL VOC validation, IoU 0.5 | mAP (%): 83.5 -> 82.9 | Hurt |
| replace the standard FPN-DCN classifier head with the RepMet DML head | object detection · ImageNet-LOC validation, IoU 0.5 | mAP (%): 55.2 -> 61.7 | Helped |
| vary the number of representatives per class and embedding architecture (ablation) | RepMet classifier · ImageNet Attributes classification | classification error standard deviation | Helped |
| use the softer normalized sum over mixture modes instead of the maximum-mode posterior (ablation) | RepMet DML classification · fine-grained classification | test error | Helped |
Closest papers by meaning
- Restoring Negative Information in Few-Shot Object Detection (NeurIPS 2020 · similarity 0.88)
- Distance Metric Learning with Joint Representation Diversification (ICML 2020 · similarity 0.86)
- Accurate Few-Shot Object Detection With Support-Query Mutual Guidance and Hybrid Loss (CVPR 2021 · similarity 0.86)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.86)
- Frustratingly Simple Few-Shot Object Detection (ICML 2020 · similarity 0.85)
- Few-Shot Object Detection via Association and DIscrimination (NeurIPS 2021 · 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.