Research map / Detection and segmentation / Open-set and few-shot detection
DiGeo: Discriminative Geometry-Aware Learning for Generalized Few-Shot Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| Replace online-trained classifier weights with fixed offline ETF weights (ablation) | DiGeo · VOC 1-shot and 5-shot | nAP50: 33.1 -> 35.8; 55.9 -> 56.4 | Helped |
| Use fixed ETF classifier weights rather than a learned classifier (ablation) | GFSOD with full base and novel training set · VOC 1-shot | nAP50: 12.2 -> 35.8 | Helped |
| Add class-specific prior margins to tighten features around class centers (ablation) | DiGeo · VOC novel classes | nAP50 | Helped |
| Up-sample images containing novel-class annotations (ablation) | DiGeo · VOC novel classes | nAP50 | Helped |
| Combine novel-image up-sampling with DiGeo (ablation) | DiGeo · VOC novel classes | nAP50 | Helped |
| Learn margins directly from scratch for each class (ablation) | DiGeo · VOC novel classes | nAP50 | Hurt |
| Share margins within base, novel, and background class groups (ablation) | DiGeo · VOC novel classes | nAP50 | Helped |
| Adapt prior-initialized margins by direct learning without self-distillation (ablation) | DiGeo · VOC | detection precision | No clear effect |
| Add a base-detector initialization stage before DiGeo prior-margin training (ablation) | DiGeo prior-margin baseline · VOC and COCO | detection precision | No clear effect |
| Share classification and localization features instead of decoupling them (ablation) | DiGeo · VOC | AP50: 74.0 | Hurt |
| Use 5 fixed background centers instead of 1 (ablation) | DiGeo · VOC 5-shot | nAP50: 56.4 -> 50.2 | Hurt |
| Use more learnable background centers (ablation) | DiGeo · VOC 5-shot | nAP50: 56.4 -> 56.7; 57.0; 55.2 | Mixed |
Closest papers by meaning
- Few-Shot Object Detection via Association and DIscrimination (NeurIPS 2021 · similarity 0.90)
- Beyond Max-Margin: Class Margin Equilibrium for Few-Shot Object Detection (CVPR 2021 · similarity 0.88)
- Restoring Negative Information in Few-Shot Object Detection (NeurIPS 2020 · similarity 0.88)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.86)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 2024 · similarity 0.86)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.86)
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.