Research map / Detection and segmentation / Open-set and few-shot detection
Restoring Negative Information in Few-Shot Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| add negative representatives and negative-aware inference (NP-RepMet) | RepMet · ImageNet-LOC 1-shot | mAP: 56.9 -> 68.5 | Helped |
| add negative representatives and negative-aware inference (NP-RepMet) | RepMet · PASCAL VOC novel set 2 1-shot | mAP: 17.2 -> 41.6 | Helped |
| add NP-RepMet at 10-shot | RepMet · PASCAL VOC novel set 1 10-shot | mAP: 49.4 vs 51.5 Meta R-CNN | Mixed |
| use two embeddings (NP-embedding) instead of one (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 65.8 -> 68.5 | Helped |
| use negative and positive representatives at inference instead of positive only (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 57.4 -> 68.5 | Helped |
| add negative representatives at inference only to RepMet-trained model (ablation) | RepMet · ImageNet-LOC 1-shot | mAP: 56.9 -> 59.4 | Helped |
| select hard negatives by spectral clustering with centroid (Cluster-Min) instead of random (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 66.5 -> 68.5 | Helped |
| choose centroid instead of random within cluster (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 67.1 -> 68.5 | Helped |
| use negative distance term (beta 0.3 vs 0) (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 54.3 -> 68.5 | Helped |
| raise negative IoU interval from (0.2,0.3) to (0.3,0.4) (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 68.5 -> 58.6 | Hurt |
| lower negative IoU interval to (0, 0.1) (ablation) | NP-RepMet · ImageNet-LOC 1-shot | mAP: 68.5 -> 68.0 | No clear effect |
Closest papers by meaning
- Few-Shot Object Detection With Attention-RPN and Multi-Relation Detector (CVPR 2020 · similarity 0.90)
- Generalized Few-Shot Object Detection Without Forgetting (CVPR 2021 · similarity 0.89)
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.89)
- Few-Shot Object Detection via Classification Refinement and Distractor Retreatment (CVPR 2021 · similarity 0.89)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · 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.