Research map / Detection and segmentation / Open-set and few-shot detection
Rethinking Few-Shot Object Detection on a Multi-Domain Benchmark
Kibok Lee, Hao Yang, Satyaki Chakraborty, Zhaowei Cai, Gurumurthy Swaminathan, Avinash Ravichandran, Onkar Dabeer
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 |
|---|---|---|---|
| tune whole network instead of detection head only (FSCE-base) (ablation) | Faster R-CNN · MoFSOD 1-shot | AP50: outperforms FSCE-base by 1.0% | Improved |
| tune whole network instead of last FC layers only (TFA) (ablation) | Faster R-CNN · MoFSOD 1-shot | AP50: 8.1% | Improved |
| tune whole network instead of detection head / last FC layers (ablation) | Faster R-CNN · MoFSOD 10-shot | AP50: 2.4%, 17.4% | Improved |
| tune last FC layers only on datasets close to COCO (KITTI, iWildCam) (ablation) | TFA · MoFSOD 5-shot per dataset | AP50: 71.7 vs 63.3 iWildCam; 48.3 vs 45.3 KITTI | Mixed |
| replace Faster R-CNN with RetinaNet (ablation) | FT · MoFSOD | AP50: 4--6% low performance | Hurt |
| Cascade R-CNN instead of Faster R-CNN (ablation) | FT · MoFSOD | AP50: 31.5 -> 31.5 (1-shot) | No clear effect |
| add FPN-P67 to Cascade R-CNN (ablation) | FT · MoFSOD | AP50: 3--4% | Improved |
| CenterNet2 instead of Faster R-CNN (ablation) | FT · MoFSOD 1-shot | AP50: lower, e.g. 2.4% in 1-shot | Hurt |
| Deformable-DETR instead of Faster R-CNN (ablation) | FT · MoFSOD 10-shot | AP50: 2.1% on MoFSOD in 10-shot | Improved |
| pre-train on ImageNet instead of COCO (ablation) | Cascade R-CNN-P67 · MoFSOD 5-shot | AP50: 49.9 -> 29.5 | Hurt |
| pre-train on Unified or LVIS instead of COCO (ablation) | Cascade R-CNN-P67 · MoFSOD 5-shot | AP50: 49.9 -> 49.6 | No clear effect |
| initialize classifier with CLIP text embeddings (LVIS+) (ablation) | CenterNet2 · MoFSOD 5-shot | AP50: 7.5% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DeFRCN — DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (ICCV 2021) · better on 5 of 5 shared results
- FSCE-base — FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding (CVPR 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
- Multi-Scale Positive Sample Refinement for Few-Shot Object Detection (ECCV 2020 · similarity 0.91)
- Frustratingly Simple Few-Shot Object Detection (ICML 2020 · similarity 0.91)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.90)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.90)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.90)
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.
