Research map / Detection and segmentation / Open-set and few-shot detection
Generalized Few-Shot Object Detection Without Forgetting
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 |
|---|---|---|---|
| Use Retentive R-CNN instead of TFA w/ cos | Few-shot object detection · MS-COCO 10-shot | AP: 28.7 -> 32.1 | Helped |
| Add max ensembling of pretrained and finetuned RPN objectness (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 32.0 -> 32.1 | Helped |
| Use max rather than geometric-average RPN ensembling (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 30.5 -> 32.1 | Helped |
| Use max rather than arithmetic-average RPN ensembling (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 32.0 -> 32.1 | No clear effect |
| Unfreeze RPN box regression layers (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 32.1 -> 32.0 | No clear effect |
| Use KL divergence rather than L1 consistency loss (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 32.0 -> 32.1 | No clear effect |
| Use KL divergence rather than cosine consistency loss (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 31.9 -> 32.1 | Helped |
| Train the adapted detector on base and novel classes rather than novel classes only (ablation) | Retentive R-CNN · MS-COCO 10-shot | nAP: 8.7 -> 10.5 | Helped |
| Unfreeze only classification and box-regression output layers rather than also unfreezing box-head linear layers (ablation) | Retentive R-CNN · MS-COCO 10-shot | AP: 31.9 -> 32.1 | Helped |
| Use a cosine classifier rather than an FC classifier (ablation) | Retentive R-CNN · MS-COCO 10-shot | nAP: 9.9 -> 10.5 | Helped |
Closest papers by meaning
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.90)
- Restoring Negative Information in Few-Shot Object Detection (NeurIPS 2020 · similarity 0.89)
- Dynamic Few-Shot Visual Learning Without Forgetting (CVPR 2018 · similarity 0.89)
- Few-Shot Object Detection With Attention-RPN and Multi-Relation Detector (CVPR 2020 · similarity 0.89)
- NIFF: Alleviating Forgetting in Generalized Few-Shot Object Detection via Neural Instance Feature Forging (CVPR 2023 · similarity 0.89)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 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.