Research map / Detection and segmentation / Open-set and few-shot detection
Towards Hard-Positive Query Mining for DETR-Based Human-Object Interaction Detection
Xubin Zhong, Changxing Ding, Zijian Li, Shaoli Huang
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 |
|---|---|---|---|
| add GBS (ground-truth box shifting hard-positive queries) (ablation) | QPIC · HICO-DET (DT mode, Full) | mAP: 29.07 -> 30.57 | Improved |
| add AMM (attention map masking) (ablation) | QPIC · HICO-DET (DT mode, Full) | mAP: 29.07 -> 30.58 | Improved |
| add GBS on V-COCO (ablation) | QPIC · V-COCO | mAP: 61.80 -> 63.19 | Improved |
| add AMM on HOI-A (ablation) | QPIC · HOI-A | mAP: 74.10 -> 75.30 | Improved |
| combine GBS and AMM by cascading (CJL) instead of using either alone (ablation) | QPIC + GBS + AMM · HICO-DET (DT mode, Full) | mAP: 30.57 -> 30.11 | Hurt |
| combine GBS and AMM in parallel (PJL) (ablation) | QPIC + GBS + AMM · HICO-DET (DT mode, Full) | mAP: 29.07 -> 30.81 | Improved |
| use alternate joint learning (AJL) of GBS and AMM instead of parallel (ablation) | QPIC + HQM · HICO-DET (DT mode, Full) | mAP: 30.81 -> 31.34 | Improved |
| shift the GT boxes in GBS instead of using unshifted GT boxes (ablation) | QPIC + GBS · HICO-DET (DT mode, Full) | mAP: 29.61 -> 30.57 | Improved |
| shift boxes in GBS instead of adding Gaussian noise (ablation) | QPIC + GBS · HICO-DET (DT mode, Full) | mAP: 30.05 -> 30.57 | Improved |
| mask top-K attention elements instead of random elements (ablation) | QPIC + AMM · HICO-DET (DT mode, Full) | mAP: 30.06 -> 30.58 | Improved |
| select masked elements from learnable query map A^l (ablation) | QPIC + AMM · HICO-DET (DT mode, Full) | mAP: 30.11 -> 30.58 | Improved |
| apply masking to separate copied queries Q^m instead of learnable queries (ablation) | QPIC + AMM · HICO-DET (DT mode, Full) | mAP: 28.75 -> 30.58 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HOTR — HOTR: End-to-End Human-Object Interaction Detection With Transformers (CVPR 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- QueryCraft: Transformer-Guided Query Initialization for Enhanced Human-Object Interaction Detection (AAAI 2026 · similarity 0.92)
- QPIC: Query-Based Pairwise Human-Object Interaction Detection With Image-Wide Contextual Information (CVPR 2021 · similarity 0.92)
- HOTR: End-to-End Human-Object Interaction Detection With Transformers (CVPR 2021 · similarity 0.90)
- Discovering Human Interactions With Large-Vocabulary Objects via Query and Multi-Scale Detection (ICCV 2021 · similarity 0.90)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.89)
- Orchestrating the Symphony of Prompt Distribution Learning for Human-Object Interaction Detection (AAAI 2025 · similarity 0.88)
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.
