Research map / Detection and segmentation / CNN features
Bounding-box Channels for Visual Relationship Detection
Sho Inayoshi, Keita Otani, Antonio Tejero-de-Pablos, Tatsuya Harada
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 bounding-box channels (vs CAI-like baseline without them) | CAI-style VRD network · VRD phrase detection, k=70 (R-CNN proposals) | R@100: 19.24 -> 28.38 | Improved |
| add bounding-box channels (vs CAI-like baseline without them) | CAI-style VRD network · VRD relationship detection, k=70 (R-CNN proposals) | R@100: 17.39 -> 22.86 | Improved |
| use Faster R-CNN proposals instead of R-CNN proposals [13] | BBCNet · VRD phrase detection, k=10 | R@50: 24.46 -> 36.71 | Improved |
| use Faster R-CNN proposals instead of R-CNN proposals [13] | BBCNet · VRD relationship detection, k=10 | R@50: 19.90 -> 30.61 | Improved |
| keep 10 predicates per object pair instead of 1 | BBCNet · VRD phrase detection (R-CNN proposals) | R@50: 19.72 -> 24.46 | Improved |
| use word2vec semantic features instead of posterior-probability features (zero-shot) | BBCNet vs VRD-Full · VRD zero-shot phrase detection, k=10 | R@50: 7.56 -> 12.57 | Improved |
| use bounding-box channels with ground-truth boxes (predicate detection) on zero-shot triplets | BBCNet vs LK distillation · VRD predicate detection zero-shot, k=1 | R@100/50: 19.17 -> 27.54 | Improved |
| BBCNet versus LSVR on relationship detection at k=1 (R-CNN proposals) | BBCNet · VRD relationship detection, k=1 | R@50: 16.08 -> 15.87 | Hurt |
| BBCNet versus RelDN at the lowest recall cutoff | BBCNet · Visual Genome SGDET (graph constraint) | R@20: 21.1 -> 20.4 | Hurt |
| BBCNet versus RelDN on scene graph detection | BBCNet · Visual Genome SGDET (graph constraint) | R@100: 32.7 -> 34.3 | Improved |
Closest papers by meaning
- MR-NET: Exploiting Mutual Relation for Visual Relationship Detection (AAAI 2019 · similarity 0.88)
- Zoom-Net: Mining Deep Feature Interactions for Visual Relationship Recognition (ECCV 2018 · similarity 0.88)
- Scene-Graph ViT: End-to-End Open-Vocabulary Visual Relationship Detection (ECCV 2024 · similarity 0.87)
- Visual Relationship Detection with Language Priors (ECCV 2016 · similarity 0.87)
- Relationship Proposal Networks (CVPR 2017 · similarity 0.87)
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.87)
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.
