Research map / Detection and segmentation / Open-set and few-shot detection
Generalized Semantic Contrastive Learning via Embedding Side Information for Few-Shot Object Detection
Ruoyu Chen, Hua Zhang, Jingzhi Li, Li Liu, Zhen Huang, Xiaochun Cao
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 |
|---|---|---|---|
| add visual-attribute knowledge matrix (ablation) | Ours (no aug.) · PASCAL VOC splits 1-3, all shots | AP50: 0.6-2.6 (an average of 1.4) improvement | Improved |
| add prototype clustering after knowledge matrix (ablation) | Ours · PASCAL VOC splits 1-3 | AP50: an average of 0.5 | Improved |
| add counterfactual data augmentation (ablation) | Ours · PASCAL VOC splits 1-3 | AP50: average performance improvement of 0.8 | Improved |
| add knowledge matrix to FSCE (ablation) | FSCE · PASCAL VOC split 1 10-shot | AP50: 63.4 -> 62.5 | Hurt |
| add knowledge matrix to FSCE (ablation) | FSCE · PASCAL VOC split 2 10-shot | AP50: 48.7 -> 51.1 | Improved |
| add prototype clustering to FSCE (ablation) | FSCE · PASCAL VOC split 1 10-shot | AP50: 63.4 -> 62.1 | Hurt |
| use self-constructed knowledge matrix instead of visual-attribute matrix (ablation) | Ours · PASCAL VOC split 1 10-shot | AP50: 64.8 -> 60.2 | Hurt |
| use visual attributes instead of text embeddings for knowledge matrix (ablation) | Ours · PASCAL VOC split 1 | AP50: 0.6 and 0.3 better performance on 1-shot and 10-shot | Improved |
| use three cluster centers instead of one (ablation) | Ours with visual matrix · PASCAL VOC split 1 10-shot | AP50: 64.84 -> 64.21 | Hurt |
| add prototype clustering with one center (ablation) | Ours with visual matrix · PASCAL VOC split 1 10-shot | AP50: 64.42 -> 64.84 | Improved |
| increase contrastive temperature from 0.2 to 1 (ablation) | Ours · PASCAL VOC split 1 10-shot | AP50: 65.2 -> 63.2 | Hurt |
| vary lambda3 from 1 to 0.5, 1.5, 2 (ablation) | Ours · PASCAL VOC split 1 10-shot | AP50: 65.2 vs 64.6, 64.5, 64.7 | No clear effect |
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.
- FSCE* — FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding (CVPR 2021) · better on 4 of 4 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
- Learning General and Specific Embedding with Transformer for Few-Shot Object Detection (IJCV 2024 · similarity 0.91)
- Fixing Background Misclassification in Few-Shot Object Detection via Product of Experts (TPAMI 2025 · similarity 0.91)
- FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding (CVPR 2021 · similarity 0.91)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.91)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.90)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.89)
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.
