Research map / Detection and segmentation / Open-set and few-shot detection
FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding
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 |
|---|---|---|---|
| add CPE supervised contrastive loss to the fine-tuned RoI head (ablation) | FSCE with RPN and RoI refinement · PASCAL VOC Novel Split 1, 3, 5, and 10 shots | Novel AP50: 49.7 -> 50.6; 58.6 -> 60.7; 61.4 -> 62.7 | Helped |
| use proposal consistency control with high-IoU proposals (ablation) | FSCE with CPE loss · PASCAL VOC Novel Split 1, 3, 5, and 10 shots | Novel AP50: 50.6 -> 51.4; 60.7 -> 61.9; 62.7 -> 63.4 | Mixed |
| fine-tune the RoI feature extractor in addition to FPN and RPN (ablation) | FSCE strong baseline · PASCAL VOC Novel Split 1, 3, 10 shots | Novel AP50: 47.2 -> 49.7; 59.8 -> 61.4 | Helped |
| increase the contrastive head dimension from 128 to 256 (ablation) | FSCE · PASCAL VOC Novel Split 1, 10 shots, temperature 0.2 | Novel AP50: 63.4 -> 63.4 | No clear effect |
| use temperature 0.2 rather than 0.07 or 0.5 (ablation) | FSCE with contrast head dimension 128 · PASCAL VOC Novel Split 1, 10 shots | Novel AP50: 63.1 -> 63.4; 62.9 -> 63.4 | Helped |
| use a hard IoU cutoff of 0.7 rather than 0.5 (ablation) | FSCE proposal consistency control · PASCAL VOC Novel Split 1, 3, 5, and 10 shots | Novel AP: 50.5 -> 50.8; 60.7 -> 61.9; 62.1 -> 63.4 | Mixed |
| weight all proposals linearly by IoU instead of applying a high-IoU cutoff (ablation) | FSCE proposal consistency control · PASCAL VOC Novel Split 1, 3, 5, and 10 shots | Novel AP: 50.8 -> 51.4; 61.9 -> 59.7; 63.4 -> 61.1 | Mixed |
| use exponential rather than linear IoU weighting (ablation) | FSCE proposal consistency control · PASCAL VOC Novel Split 1, 3, 5, and 10 shots | Novel AP: 51.4 -> 50.8; 59.7 -> 59.6; 61.1 -> 61.6 | Hurt |
| add contrastive proposal encoding to few-shot detection | Faster R-CNN R101-FPN · PASCAL VOC Novel Split 1, all reported shots | nAP50: TFA w/ cos: 39.8, 36.1, 44.7, 55.7, 56.0; FSCE: 44.2, 43.8, 51.4, 61.9, 63.4 | Helped |
| use FSCE rather than TFA w/ cos | Faster R-CNN R101-FPN · COCO novel classes, 10 and 30 shots | AP: 10.0 -> 11.9; 13.7 -> 16.4 | Helped |
| use FSCE rather than the strongest listed baseline | Faster R-CNN R101-FPN · COCO novel classes, 10 and 30 shots | AP75: 9.8 -> 10.5; 14.2 -> 16.2 | Helped |
Closest papers by meaning
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.91)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.90)
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.90)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.89)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 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.