Research map / Video / Action recognition
Temporal Segment Networks: Towards Good Practices for Deep Action Recognition
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 temporal segment network (sparse sampling, video-level supervision) (ablation) | BN-Inception two-stream · UCF101 split 1 | accuracy: 92.0% -> 93.5% | Improved |
| add cross modality pre-training (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 |
| pre-train from ImageNet instead of training from scratch (ablation) | two-stream ConvNets (spatial) · UCF101 split 1 | spatial accuracy: 48.7% -> 84.1% | Improved |
| use average instead of max consensus (ablation) | TSN · UCF101 split 1 | two-stream accuracy: 91.6% -> 93.5% | Improved |
| use average instead of weighted average consensus (ablation) | TSN · UCF101 split 1 | two-stream accuracy: 92.4% -> 93.5% | Improved |
| use BN-Inception instead of VGGNet-16 (ablation) | two-stream ConvNets · UCF101 split 1 | two-stream accuracy: 90.9% -> 92.0% | Improved |
| use BN-Inception instead of Clarifai (ablation) | two-stream ConvNets · UCF101 split 1 | two-stream accuracy: 87.0% -> 92.0% | Improved |
| add RGB difference to RGB image (ablation) | spatial stream · UCF101 split 1 | accuracy: 84.5% -> 87.3% | Improved |
| add warped optical flow to optical flow (ablation) | temporal stream · UCF101 split 1 | accuracy: 87.2% -> 87.8% | Improved |
| add RGB difference to optical flow + warped flow + RGB (ablation) | multi-modality TSN · UCF101 split 1 | accuracy: 92.3% -> 91.7% | Hurt |
| add warped flow as third modality to TSN (ablation) | TSN · HMDB51 (3 splits average) | accuracy: 68.5% -> 69.4% | Improved |
Closest papers by meaning
- Temporal Segment Networks for Action Recognition in Videos (TPAMI 2018 · similarity 0.97)
- MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation (CVPR 2019 · similarity 0.93)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.93)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.91)
- Temporal Pyramid Network for Action Recognition (CVPR 2020 · similarity 0.90)
- 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.
