Research map / Video / Action recognition
TDN: Temporal Difference Networks for Efficient Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 temporal difference to both S-TDM and L-TDM instead of concatenating frames and averaging (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 41.5% -> 52.3% | Helped |
| use additive fusion rather than attention-based fusion in S-TDM (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 43.7% -> 52.3% | Helped |
| use spatiotemporal attention rather than direct addition in L-TDM (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 44.1% -> 52.3% | Helped |
| use multi-scale difference extraction in L-TDM (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 49.7% -> 52.3% | Helped |
| use bidirectional rather than one-directional difference in L-TDM (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 50.0% -> 52.3% | Helped |
| place S-TDM in Stages 1-2 and L-TDM in Stages 3-5 (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 46.6% -> 52.3% | Helped |
| add S-TDM and L-TDM together rather than using either alone (ablation) | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 51.5% -> 52.3%; 49.9% -> 52.3% | Helped |
| use TDM instead of temporal convolution, TSM, or TEINet with 8 frames | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 47.5% -> 52.3%; 47.1% -> 52.3%; 48.4% -> 52.3% | Helped |
| increase TDN input from 8 frames to 16 frames | TDN-ResNet50 · Something-Something V1 | Top1 accuracy: 52.3 -> 53.9 | Helped |
| increase TDN input from 8 frames to 16 frames | TDN-ResNet50 · Something-Something V2 | Top1 accuracy: 64.0 -> 65.3 | Helped |
| increase TDN input from 8 frames to 16 frames | TDN-ResNet50 · Kinetics-400 validation | Top1 accuracy: 76.6 -> 77.5 | Helped |
Closest papers by meaning
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.90)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.89)
- Temporal Deformable Residual Networks for Action Segmentation in Videos (CVPR 2018 · similarity 0.89)
- Temporal Pyramid Network for Action Recognition (CVPR 2020 · similarity 0.89)
- MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation (CVPR 2019 · similarity 0.89)
- Memory-Augmented Temporal Dynamic Learning for Action Recognition (AAAI 2019 · similarity 0.88)
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.