Research map / Detection and segmentation / Open-set and few-shot detection
Pairwise Body-Part Attention for Recognizing Human-Object Interactions
Hao-Shu Fang, Jinkun Cao, Yu-Wing Tai, Cewu Lu
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 object features (union box) (ablation) | baseline model · HICO test | mAP: 35.6 -> 37.0 | Improved |
| use union box instead of tight box for object features (ablation) | baseline model · HICO test | mAP: 36.3 -> 37.0 | Improved |
| add body-part features without attention (ablation) | baseline + object · HICO test | mAP: 37.0 -> 38.0 | Improved |
| use body-part pairs instead of single body parts (no attention) (ablation) | body-part model · HICO test | mAP: 38.0 -> 38.9 | Improved |
| add pairwise body-part attention with feature selection (ablation) | body-part pairs model · HICO test | mAP: 38.9 -> 39.9 | Improved |
| attend over both single parts and pairs (ablation) | pairwise attention model · HICO test | mAP: 39.9 -> 39.1 | Hurt |
| use feature selection (k=20) instead of keeping all 45 pairs (ablation) | pairwise attention model · HICO val | mAP: 1.2 mAP higher than k=45 | Improved |
| add weighted sigmoid cross entropy loss | Ours · HICO test | mAP: 37.5 -> 39.9 | Improved |
| replace VGG16 with ResNet101 base network | Ours · MPII test | mAP: 36.8 -> 37.5 | Improved |
| use pairwise body-part attention vs Mallya & Lazebnik (weighted loss) | Mallya & Lazebnik · HICO test | mAP: 36.1 -> 39.9 | Improved |
| use pairwise body-part attention on apple buying class | Mallya & Lazebnik · HICO test, class: apple buying | AP: 19.3 -> 59.0 | Improved |
| use pairwise body-part attention on oven cleaning class | Mallya & Lazebnik · HICO test, class: oven cleaning | AP: 22.1 -> 13.1 | Hurt |
Closest papers by meaning
- Mining Cross-Person Cues for Body-Part Interactiveness Learning in HOI Detection (ECCV 2022 · similarity 0.90)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.87)
- Exploring the Potential of Large Foundation Models for Open-Vocabulary HOI Detection (CVPR 2024 · similarity 0.87)
- HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · similarity 0.87)
- Learning Human-Object Interaction as Groups (NeurIPS 2025 · similarity 0.86)
- PairDETR : Joint Detection and Association of Human Bodies and Faces (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.
