Research map / Detection and segmentation / Open-set and few-shot detection
Mining the Benefits of Two-stage and One-stage HOI Detection
Aixi Zhang, Yue Liao, Si Liu, Miao Lu, Yongliang Wang, Chen Gao, XIAOBO LI
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Replace coupled HOI prediction with cascade-disentangled human-object detection and interaction classification (ablation) | QPIC-style one-stage HOI detection · HICO-Det, Full | mAP: 29.07 -> 31.06 | Improved |
| Add re-weighted training to the cascade-disentangled base model (ablation) | CDN-B · HICO-Det, Full | mAP: 31.06 -> 31.38 | Improved |
| Add re-weighted training to the cascade-disentangled base model (ablation) | CDN-B · HICO-Det, Rare | mAP: 26.68 -> 27.36 | Improved |
| Add PNMS to the re-weighted model (ablation) | CDN-B · HICO-Det, Full | mAP: 31.38 -> 31.78 | Improved |
| Decouple training without re-weighting (ablation) | CDN-B base model · HICO-Det, Full | mAP: 31.06 -> 30.90 | Hurt |
| Add static class weights (ablation) | CDN-B · HICO-Det, Full | mAP: 30.90 -> 31.25 | Improved |
| Use dynamic instead of static re-weighting (ablation) | CDN-B · HICO-Det, Full | mAP: 31.25 -> 31.38 | Improved |
| Set dynamic-weight queue length to 2×N_s and p to 0.7 (ablation) | CDN-B dynamic re-weighting · HICO-Det, Full | mAP: 31.34 -> 31.38 | Improved |
| Change dynamic-weight p from 0.7 to 0.8 at L_Q = 2×N_s (ablation) | CDN-B dynamic re-weighting · HICO-Det, Full | mAP: 31.38 -> 31.33 | Hurt |
| Set PNMS threshold to 0.7 instead of 0.8 with α = 1 and β = 1 (ablation) | CDN-B with PNMS · HICO-Det, Full | mAP: 31.66 -> 31.75 | Improved |
| Set PNMS β to 0.5 instead of 1 at threshold 0.7 (ablation) | CDN-B with PNMS · HICO-Det, Full | mAP: 31.75 -> 31.78 | Improved |
| Replace QPIC interaction prediction with an isolated interaction decoder initialized from human-object pair queries | QPIC · HICO-Det, Full | mAP: 29.07 -> 30.96 | Improved |
Closest papers by meaning
- Diagnosing Human-Object Interaction Detectors (IJCV 2025 · similarity 0.92)
- Learning Human-Object Interaction Detection Using Interaction Points (CVPR 2020 · similarity 0.91)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.89)
- Exploring the Potential of Large Foundation Models for Open-Vocabulary HOI Detection (CVPR 2024 · similarity 0.89)
- HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · similarity 0.89)
- HOIMamba: Efficient Mamba-based Disentangled Progressive Learning for HOI Detection (AAAI 2025 · similarity 0.89)
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.
