Research map / Detection and segmentation / Open-set and few-shot detection
Adaptive Multi-task Learning for Few-shot Object Detection
Yan Ren, Yanling Li, Adams Wai-Kin Kong
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 |
|---|---|---|---|
| split shared RCNN head into double-head classification/localization branches (ablation) | DeFRCN baseline · COCO 10-shot | mAP: 16.7 -> 17.4 | Improved |
| add precision-driven gradient balancer to double-head RCNN (ablation) | Enhanced baseline · COCO 10-shot | mAP: 17.4 -> 18.6 | Improved |
| use PCGrad as gradient modulation instead of the proposed balancer (ablation) | Enhanced baseline · COCO 10-shot | mAP: 17.4 -> 9.8 | Hurt |
| use existing closed-form gradient methods (GradVac, NashMTL) (ablation) | Enhanced baseline · COCO 10-shot | mAP: 17.4 -> 17.3 (GradVac), 17.2 (NashMTL) | No clear effect |
| add CLIP knowledge distillation (ablation) | Double-head + balancer · COCO 10-shot | mAP: 18.4 -> 18.6 | Improved |
| add CLIP-based score refinement at inference (ablation) | Full method · COCO 10-shot | mAP: 18.6 -> 22.7 | Improved |
| add score refinement to the DeFRCN baseline instead of PCB (ablation) | DeFRCN · COCO 10-shot | mAP: 18.5 -> 19.6 | Improved |
| tune score refinement trade-off omega (ablation) | Score refinement · COCO 10-shot | mAP: 18.50 (no refine) -> 22.7 (omega 0.6) | Improved |
| adaptive balancer vs grid-searched gradient ratios (ablation) | Gradient balancer · COCO 10-shot | gradient ratios | No clear effect |
| use too small gamma_cls and gamma_reg (ablation) | Gradient balancer · COCO 10-shot | mAP: 0.5 (gamma 0.5) vs 18.4 (gamma 1) | Hurt |
| add full method on cross-domain COCO to VOC | DeFRCN · COCO to VOC cross-domain | mAP: 41.0 -> 42.5 | Improved |
| add full method vs RISF on COCO mid-shots | Full method · COCO 3-shot and 5-shot | mAP: 17.5 vs 18.2 (3-shot), 19.5 vs 20.3 (5-shot) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeFRCN — DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (ICCV 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Meta-Learning to Detect Rare Objects (ICCV 2019 · similarity 0.90)
- Frustratingly Simple Few-Shot Object Detection (ICML 2020 · similarity 0.88)
- Few-Shot Object Detection via Feature Reweighting (ICCV 2019 · similarity 0.88)
- Multi-Scale Positive Sample Refinement for Few-Shot Object Detection (ECCV 2020 · similarity 0.87)
- TOOD: Task-Aligned One-Stage Object Detection (ICCV 2021 · similarity 0.87)
- 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.
