Research map / Detection and segmentation / Open-set and few-shot detection
Few-Shot Object Detection via Feature Reweighting
Bingyi Kang, Zhuang Liu, Xin Wang, Fisher Yu, Jiashi Feng, Trevor Darrell
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 |
|---|---|---|---|
| apply reweighting at layer 21 instead of layer 13 (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 40.7 -> 47.2 | Improved |
| apply reweighting at layer 21 instead of layer 20 (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 43.6 -> 47.2 | Improved |
| reweight all layer 21 features instead of half (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 46.9 -> 47.2 | No clear effect |
| use softmax loss instead of single-binary loss (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 14.8 -> 47.2 | Improved |
| use softmax loss instead of multi-binary loss (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 41.6 -> 47.2 | Improved |
| add the mask channel to image input of reweighting module (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 43.3 -> 47.2 | Improved |
| add cropped object input on top of image+mask (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 47.2 -> 46.8 | No clear effect |
| average reweighting vectors of k-shot examples at test time instead of random single example (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 45.1 -> 47.2 | Improved |
| do not share weights between feature extractor and reweighting module (ablation) | feature reweighting detector · PASCAL VOC novel set 1, 10-shot | novel mAP: 44.8 -> 47.2 | Improved |
| two-phase training (base training then few-shot fine-tuning) instead of joint training | YOLOv2 baseline · PASCAL VOC novel set 1, 10-shot | novel mAP: 1.8 -> 38.6 | Improved |
| add feature reweighting module to YOLOv2 few-shot fine-tuning | YOLOv2 · PASCAL VOC novel set 1, 1-shot | novel mAP: 6.6 -> 14.8 | Improved |
| add LSTD background depression and transfer knowledge to YOLO-ft-full | YOLO-ft-full · PASCAL VOC novel set 2, 5-shot | novel mAP: -5.1 mAP | Hurt |
Closest papers by meaning
- Meta-Learning to Detect Rare Objects (ICCV 2019 · similarity 0.93)
- Generalized Few-Shot Object Detection Without Forgetting (CVPR 2021 · similarity 0.91)
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.91)
- Frustratingly Simple Few-Shot Object Detection (ICML 2020 · similarity 0.91)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
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.
