Research map / Detection and segmentation / Open-set and few-shot detection
Semantic Relation Reasoning for Shot-Stable Few-Shot Object Detection
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 semantic space projection to Faster R-CNN (ablation) | Faster R-CNN on VOC Novel Set 1 · 1-shot | mAP50: 32.6 -> 40.5 | Helped |
| Add semantic space projection to Faster R-CNN (ablation) | Faster R-CNN on VOC Novel Set 1 · 10-shot | mAP50: 55.6 -> 52.2 | Hurt |
| Add dynamic relation reasoning to semantic space projection (ablation) | Faster R-CNN on VOC Novel Set 1 · 1-shot | mAP50: 40.5 -> 44.1 | Helped |
| Add decoupled fine-tuning to SSP + RR (ablation) | Faster R-CNN on VOC Novel Set 1 · 1-shot | mAP50: 44.1 -> 47.8 | Helped |
| Replace the dynamic relation graph with independent trainable embedding transformation (ablation) | SSP on VOC Novel Set 1 · 1-shot | mAP50: 44.1 -> 39.3 | Hurt |
| Replace the dynamic relation graph with a heuristic knowledge graph (ablation) | SSP on VOC Novel Set 1 · 1-shot | mAP50: 44.1 -> 41.6 | Hurt |
| Use meaningful Word2Vec embeddings rather than random vectors (ablation) | Semantic space projection on VOC Novel Set 1 · 1-shot | mAP50: 33.2 -> 42.8 | Helped |
| Replace Word2Vec-300d with GloVe-300d (ablation) | Semantic space projection on VOC Novel Set 1 · 1-shot | mAP50: 42.8 -> 38.8 | Hurt |
| Vary relation feature dimension from 128 to 16 (ablation) | Relation reasoning on VOC Novel Set 1 · 1-shot | mAP50: 40.9; 42.0; 42.4; 44.1 | Mixed |
| Unfreeze FCs, RPN, and backbone during fine-tuning (ablation) | TFA on VOC Novel Set 1 · 1-shot | mAP50: 39.8 -> 16.2 | Hurt |
| Use SRR-FSD instead of TFA | Few-shot object detection on VOC Novel Set 1 · 1-shot | mAP50: 39.8 -> 47.8 | Helped |
| Use SRR-FSD instead of Meta R-CNN | Few-shot object detection on COCO novel classes · 10-shot | AP: 8.7 -> 11.3 | Helped |
Closest papers by meaning
- Stabilizing Few-Shot Object Detection with Language-Conditioned Probabilistic Prototypes (NeurIPS 2026 · similarity 0.89)
- One-Shot Learning for Long-Tail Visual Relation Detection (AAAI 2020 · similarity 0.89)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.88)
- Objects in Semantic Topology (ICLR 2022 · similarity 0.88)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Simple Semantic-Aided Few-Shot Learning (CVPR 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.