Research map / Video / Video-language understanding
Temporal Relational Reasoning in Videos
Bolei Zhou, Alex Andonian, Aude Oliva, Antonio Torralba
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 multi-scale temporal relation module over single-frame baseline | BN-Inception · Something-V1 val | top-1 accuracy: 11.41 -> 34.44 | Improved |
| add optical flow stream (2-stream TRN) | MultiScale TRN · Something-V1 val | top-1 accuracy: 34.44 -> 42.01 | Improved |
| use ordered temporal relations instead of average pooling (TRN vs TSN), 2 frames (ablation) | BN-Inception · Something-V1 val | top-1 accuracy: 16.72 -> 22.23 | Improved |
| use ordered temporal relations instead of average pooling (TRN vs TSN), 7 frames (ablation) | BN-Inception · Something-V1 val | top-1 accuracy: 18.48 -> 31.01 | Improved |
| add more frames in TRN relation (2 to 7 frames) (ablation) | TRN · Something-V1 val | top-1 accuracy: 22.23 -> 31.01 | Improved |
| add multi-scale relations (2 to 5 frames combined) versus single 5-frame TRN (ablation) | TRN · Jester val | top-1 accuracy: 91.40 -> 95.31 | Improved |
| increase relation frames from 2 to 5 (ablation) | TRN · Jester val | top-1 accuracy: 75.65 -> 91.40 | Improved |
| TR pool instead of average pool (ablation) | BN-Inception 8 frames · Something | accuracy: 19.53 -> 34.44 | Improved |
| TR pool instead of average pool (ablation) | BN-Inception 8 frames · Jester | accuracy: 85.41 -> 95.31 | Improved |
| TR pool instead of average pool (ablation) | BN-Inception 8 frames · Charades | accuracy: 11.32 -> 25.20 | Improved |
| TR pool instead of average pool (ablation) | BN-Inception 8 frames · UCF101 and Kinetics | accuracy: 82.69 -> 83.83; 63.34 -> 63.18 | No clear effect |
| TR pool instead of average pool (ablation) | BN-Inception 8 frames · Moments | accuracy: 24.11 -> 25.94 | Improved |
Closest papers by meaning
- Temporal Relational Modeling with Self-Supervision for Action Segmentation (AAAI 2021 · similarity 0.88)
- Temporal Recurrent Networks for Online Action Detection (ICCV 2019 · similarity 0.87)
- Unified Graph Structured Models for Video Understanding (ICCV 2021 · similarity 0.87)
- Social Relation Recognition From Videos via Multi-Scale Spatial-Temporal Reasoning (CVPR 2019 · similarity 0.86)
- CoTeRe-Net: Discovering Collaborative Ternary Relations in Videos (ECCV 2020 · similarity 0.86)
- TDN: Temporal Difference Networks for Efficient Action Recognition (CVPR 2021 · 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.
