Research map / Video / Action recognition
Temporal Distinct Representation Learning for Action Recognition
Junwu Weng, Donghao Luo, Yabiao Wang, Ying Tai, Chengjie Wang, Jilin Li, Feiyue Huang, Xudong Jiang, Junsong Yuan
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 PEM (ablation) | TIM baseline · Something-Something V1 val (8f) | top-1 accuracy: 46.1 -> 48.7 | Improved |
| add TD Loss (ablation) | TIM baseline · Something-Something V1 val (8f) | top-1 accuracy: 46.1 -> 48.4 | Improved |
| add PEM and TD Loss (ablation) | TIM baseline · Something-Something V1 val (8f) | top-1 accuracy: 46.1 -> 49.8 | Improved |
| add PEM instead of MEM channel enhancement (ablation) | TIM baseline · Something-Something V1 val (8f) | top-1 accuracy: 47.4 -> 48.7 | Improved |
| add PEM (ablation) | TSM baseline · Something-Something V1 val (8f) | top-1 accuracy: 45.6 -> 48.1 | Improved |
| add TD Loss (ablation) | TSM baseline · Something-Something V1 val (8f) | top-1 accuracy: 45.6 -> 47.5 | Improved |
| add PEM and TD Loss (ablation) | TSM baseline · Something-Something V1 val (8f) | top-1 accuracy: 45.6 -> 48.4 | Improved |
| regularize 50% of channels with TD Loss instead of 100% (ablation) | TIM + TD Loss · Something-Something V1 val (8f) | top-1 accuracy: 47.3 -> 48.4 | Improved |
| place TD Loss after temporal module instead of before (ablation) | TIM + TD Loss · Something-Something V1 val (8f) | top-1 accuracy: 46.9 -> 48.4 | Improved |
| place PEM after temporal module instead of before (ablation) | TIM + PEM · Something-Something V1 val (8f) | top-1 accuracy: 48.7 -> 49.0 | No clear effect |
| vary lambda from 1e-4 to 4e-4 (ablation) | TIM + TD Loss · Something-Something V1 val (8f) | top-1 accuracy: 47.9, 48.4, 47.8, 48.0 | No clear effect |
| add PEM (reduces training accuracy) (ablation) | TIM baseline · Something-Something V1 train | training accuracy: 79.03 -> 77.98 | Hurt |
Closest papers by meaning
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.89)
- Long-Term Temporal Convolutions for Action Recognition (TPAMI 2017 · similarity 0.88)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.87)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.87)
- Learning Comprehensive Motion Representation for Action Recognition (AAAI 2021 · similarity 0.87)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · 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.
