Research map / Video / Action recognition
Temporal Segment Networks for Action Recognition in Videos
Limin Wang, Yuanjun Xiong, Zhe Wang, Yu Qiao, Dahua Lin, Xiaoou Tang, Luc Van Gool
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 cross modality pre-training of the temporal stream (ablation) | two-stream ConvNets · UCF101 split 1 | accuracy: 90.0% -> 91.5% | Improved |
| add partial BN with dropout (ablation) | two-stream ConvNets · UCF101 split 1 | accuracy: 91.5% -> 92.0% | Improved |
| train from scratch instead of ImageNet pre-training of the spatial stream (ablation) | two-stream ConvNets · UCF101 split 1 | accuracy: 90.0% -> 82.9% | Hurt |
| add temporal segment network (K=7) sampling | two-stream RGB+Flow · UCF101 | accuracy: 92.4% -> 94.9% | Improved |
| add TSN to RGB+RGB Diff. | RGB+RGB Diff. · UCF101 | accuracy: 86.8% -> 91.0% | Improved |
| add warped optical flow as a third modality (ablation) | TSN RGB+Flow · UCF101 | accuracy: 94.9% -> 95.0% | No clear effect |
| increase segment number K from 3 to 7 (ablation) | TSN · UCF101 | accuracy: 94.2% -> 94.9% | Improved |
| increase segment number K from 7 to 9 (ablation) | TSN · UCF101 | accuracy: 94.9% -> 94.9% | No clear effect |
| replace average pooling with max pooling as consensus (ablation) | TSN K=7 · UCF101 | accuracy: 94.9% -> 92.4% | Hurt |
| replace average pooling with top-K pooling as consensus (ablation) | TSN K=7 · ActivityNet | accuracy: 87.8% -> 88.1% | Improved |
| replace BN-Inception two-stream with TSN (K=7) | BN-Inception · ActivityNet v1.3 val | mAP: 78.9% -> 84.7% | Improved |
| use Inception V3 instead of BN-Inception in TSN (ablation) | TSN · ActivityNet v1.3 val | mAP: 84.7% -> 87.7% | Improved |
Closest papers by meaning
- Temporal Segment Networks: Towards Good Practices for Deep Action Recognition (ECCV 2016 · similarity 0.97)
- MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation (CVPR 2019 · similarity 0.94)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.93)
- Temporal Action Detection With Structured Segment Networks (ICCV 2017 · similarity 0.91)
- Temporal Pyramid Network for Action Recognition (CVPR 2020 · similarity 0.91)
- TDN: Temporal Difference Networks for Efficient Action Recognition (CVPR 2021 · similarity 0.90)
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.
