Research map / Detection and segmentation / Open-set and few-shot detection
Shuffle-Then-Assemble: Learning Object-Agnostic Visual Relationship Features
Xu Yang, Hanwang Zhang, Jianfei Cai
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 OA conv-layers (no pre-training) (ablation) | Base relationship model · VRD supervised | Recall@100: 39.25 -> 43.29 | Improved |
| add OA conv-layers (no pre-training) (ablation) | Base relationship model · VG supervised | Recall@100: 52.61 -> 58.53 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VRD supervised | Recall@100: 43.29 -> 48.03 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VG supervised | Recall@100: 58.53 -> 62.94 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VRD weakly supervised | Recall@100: 31.47 -> 35.89 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VG weakly supervised | Recall@100: 47.72 -> 51.92 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VRD zero-shot | Recall@100: 16.53 -> 20.57 | Improved |
| add Shuffle-Then-Assemble pre-training to OA layers (ablation) | Base+OA · VG zero-shot | Recall@100: 13.09 -> 18.90 | Improved |
| fine-tune pre-trained OA layers instead of freezing them (ablation) | STA · VRD supervised | Recall@100: 44.30 -> 48.03 | Improved |
| freeze pre-trained OA layers versus training OA layers from scratch (ablation) | STA w/o FT vs Base+OA · VRD supervised | Recall@100: 43.29 -> 44.30 | No clear effect |
| use residual structure in the transformation network (ablation) | STA · VRD supervised | Recall@100: 46.83 -> 48.03 | Improved |
| use residual structure in the transformation network (ablation) | STA · VG supervised | Recall@100: 62.32 -> 62.94 | Improved |
Closest papers by meaning
- Tensorize, Factorize and Regularize: Robust Visual Relationship Learning (CVPR 2018 · similarity 0.89)
- Large-Scale Visual Relationship Understanding (AAAI 2019 · similarity 0.89)
- Zoom-Net: Mining Deep Feature Interactions for Visual Relationship Recognition (ECCV 2018 · similarity 0.88)
- Weakly-Supervised Learning of Visual Relations (ICCV 2017 · similarity 0.88)
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.88)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (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.
