Research map / Detection and segmentation / CNN features
Relation Networks for 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 |
|---|---|---|---|
| Add relation modules to the instance-recognition head | Faster R-CNN, ResNet-101 · COCO minival | mAP: 32.2 -> 34.7 | Helped |
| Add relation modules to the instance-recognition head | FPN, ResNet-101 · COCO minival | mAP: 36.8 -> 38.1 | Helped |
| Add relation modules to the instance-recognition head | DCN, ResNet-101 · COCO minival | mAP: 37.5 -> 38.1 | Helped |
| Train instance recognition and duplicate removal end-to-end | Faster R-CNN + relation modules · COCO minival | mAP: 30.5 -> 31.0 | Helped |
| Use pairwise geometric attention weights instead of no geometry or unary geometry (ablation) | Relation module · COCO minival | mAP: 30.3, 31.1 -> 31.9 | Helped |
| Increase relation count from 1 to 16 (ablation) | Relation module · COCO minival | mAP: 30.5 -> 31.9 | Helped |
| Increase relation-module count from {1,1} to {4,4} (ablation) | 2fc+RM head · COCO minival | mAP: 31.9 -> 32.8 | Helped |
| Replace relation modules with a wider 2fc head (ablation) | Faster R-CNN head · COCO minival | mAP: 29.6 -> 29.7 | No clear effect |
| Replace relation modules with a deeper 3fc head (ablation) | Faster R-CNN head · COCO minival | mAP: 29.6 -> 29.0 | Hurt |
| Add global context features without relation modules (ablation) | 2fc head · COCO minival | mAP: 29.6 -> 29.6 | No clear effect |
| Add larger-window context features (ablation) | 2fc head · COCO minival | mAP: 29.6 -> 30.4 | Helped |
| Remove rank feature (ablation) | Duplicate removal network · COCO minival | mAP: 30.3 -> 26.6 | Hurt |
Closest papers by meaning
- Detecting Visual Relationships With Deep Relational Networks (CVPR 2017 · similarity 0.88)
- MR-NET: Exploiting Mutual Relation for Visual Relationship Detection (AAAI 2019 · similarity 0.88)
- RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder (NeurIPS 2020 · similarity 0.87)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.87)
- Can Transformers Capture Spatial Relations between Objects? (ICLR 2024 · similarity 0.85)
- PairDETR : Joint Detection and Association of Human Bodies and Faces (CVPR 2024 · similarity 0.85)
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.