Research map / Video / Action recognition
CoTeRe-Net: Discovering Collaborative Ternary Relations in Videos
Zhensheng Shi, Cheng Guan, Liangjie Cao, Qianqian Li, Ju Liang, Zhaorui Gu, Haiyong Zheng, Bing Zheng
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 channel relations (r=1) (ablation) | 3D ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 43.1 -> 44.2 | Improved |
| add temporal relations (r=32) (ablation) | 3D ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 43.1 -> 44.5 | Improved |
| add spatial relations with 3x3 convolution instead of 1x1 (ablation) | 3D ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 43.4 -> 44.7 | Improved |
| lower MLP reduction ratio for channel relations (r=4 to r=1) (ablation) | CoTeRe-ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 43.3 -> 44.2 | Improved |
| lower MLP reduction ratio for temporal relations (r=128 to r=32) (ablation) | CoTeRe-ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 43.3 -> 44.5 | Improved |
| collaborate channel and temporal relations instead of one relation alone (ablation) | CoTeRe-ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 44.5 -> 44.9 | Improved |
| collaborate temporal and spatial relations (ablation) | CoTeRe-ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 44.7 -> 45.4 | Improved |
| add all three relations (ternary) instead of temporal+spatial (ablation) | CoTeRe-ResNet-18 · Something-Something V1 (ablation, from scratch) | top-1 accuracy: 45.4 -> 45.8 | Improved |
| combine ternary relations by addition instead of the product form of Eq. 12 (ablation) | CTSR module · Something-Something V1 (ablation, from scratch) | top-1 accuracy: about 0.3% lower | Hurt |
| add CTSR modules (CoTeRe-Net) to ResNet-34 baseline, no pretrain | ResNet-34 · Something-Something V1 | top-1 accuracy: 47.2 -> 49.7 | Improved |
| add CTSR modules (CoTeRe-Net) to ResNet-34 baseline, no pretrain | ResNet-34 · Something-Something V2 | top-1 accuracy: 60.8 -> 63.2 | Improved |
| use ResNet-34 instead of ResNet-18 backbone (Kinetics pretrain) (ablation) | CoTeRe-Net · Something-Something V1 | top-1 accuracy: 50.5 -> 52.8 | Improved |
Closest papers by meaning
- Open-Vocabulary Video Relation Extraction (AAAI 2024 · similarity 0.88)
- Bilateral Relation Distillation for Weakly Supervised Temporal Action Localization (TPAMI 2023 · similarity 0.87)
- Appearance-and-Relation Networks for Video Classification (CVPR 2018 · similarity 0.87)
- Video Modeling With Correlation Networks (CVPR 2020 · similarity 0.86)
- Graph-Based High-Order Relation Modeling for Long-Term Action Recognition (CVPR 2021 · similarity 0.86)
- Temporal Relational Reasoning in Videos (ECCV 2018 · 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.
