Research map / Detection and segmentation / Open-set and few-shot detection
RLIP: Relational Language-Image Pre-training for Human-Object Interaction Detection
Hangjie Yuan, Jianwen Jiang, Samuel Albanie, Tao Feng, Ziyuan Huang, Dong Ni, Mingqian Tang
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 object-detection pre-training with relational language-image pre-training on VG | ParSeD · HICO-DET Default, Full | mAP: 23.78 -> 29.21 | Improved |
| replace relation-detection pre-training with RLIP on VG | ParSeD · HICO-DET Default, Full | mAP: 27.45 -> 29.21 | Improved |
| initialize RLIP with COCO object-detection parameters | RLIP-ParSeD · HICO-DET Default, Full | mAP: 29.12, 29.21 -> 30.70 | Improved |
| use RLIP-ParSe instead of ParSe | ParSe · HICO-DET UC-NF, Full | mAP: 23.38 -> 26.19 | Improved |
| use RLIP pre-training rather than object-detection pre-training | ParSeD · HICO-DET, 1% few-shot, Full | mAP: 9.17 -> 16.09 | Improved |
| fully disentangle subject, object, and relation representations (ablation) | ParSe · HICO-DET fine-tuning, Full | mAP: 29.55 -> 31.79 | Improved |
| add Label Sequence Extension (ablation) | RLIP-ParSeD · HICO-DET fine-tuning, Full | mAP: 27.51 -> 28.22 | Improved |
| add Relational Quality Labels after LSE (ablation) | RLIP-ParSeD · HICO-DET fine-tuning, Full | mAP: 28.22 -> 28.94 | Improved |
| add Relational Pseudo-Labels after LSE and RQL (ablation) | RLIP-ParSeD · HICO-DET fine-tuning, Full | mAP: 28.94 -> 29.21 | Improved |
| use uniform rather than frequency-based LSE sampling (ablation) | RLIP-ParSeD · HICO-DET fine-tuning, Full | mAP: 28.22 -> 28.12 | Hurt |
| replace RPL+RQL with generic Label Smoothing Regularization (ablation) | RLIP-ParSeD with LSE · HICO-DET fine-tuning, Full | mAP: 28.22 -> 28.03 | Hurt |
| use Euclidean rather than Cosine distance for RPL at η=0.3 (ablation) | RLIP-ParSeD with LSE and RQL · HICO-DET zero-shot NF, Full | mAP: 12.23 -> 12.69 | Improved |
Closest papers by meaning
- LINK: Learning Instance-level Knowledge from Vision-Language Models for Human-Object Interaction Detection (ICLR 2026 · similarity 0.92)
- HOICLIP: Efficient Knowledge Transfer for HOI Detection With Vision-Language Models (CVPR 2023 · similarity 0.91)
- Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models (NeurIPS 2023 · similarity 0.90)
- Human-Object Interaction Detection Collaborated with Large Relation-driven Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Learning Transferable Human-Object Interaction Detector With Natural Language Supervision (CVPR 2022 · similarity 0.90)
- Discovering Syntactic Interaction Clues for Human-Object Interaction Detection (CVPR 2024 · 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.
