Research map / Video / Action recognition
STM: SpatioTemporal and Motion Encoding for Action Recognition
Boyuan Jiang, MengMeng Wang, Weihao Gan, Wei Wu, Junjie Yan
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 CSTM and CMM (STM) to TSN | TSN ResNet-50 · Something-Something v1 val (8 frames) | top-1 accuracy: 19.7 -> 49.2 | Improved |
| add CMM to CSTM (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 47.7 -> 49.2 | Improved |
| add CSTM to CMM (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 44.1 -> 49.2 | Improved |
| add CSTM alone to TSN (ablation) | TSN · Something-Something v1 val (8 frames) | top-1 accuracy: 19.7 -> 47.7 | Improved |
| add CMM alone to TSN (ablation) | TSN · Something-Something v1 val (8 frames) | top-1 accuracy: 19.7 -> 44.1 | Improved |
| fuse CSTM and CMM by summation instead of concatenation (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 41.8 -> 49.2 | Improved |
| replace only one block with STM at stage 5 instead of stage 2 (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 38.7 -> 41.5 | Improved |
| replace one block per stage (4 blocks) instead of one block (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 41.5 -> 47.9 | Improved |
| replace all 16 blocks instead of 4 (ablation) | STM · Something-Something v1 val (8 frames) | top-1 accuracy: 47.9 -> 49.2 | Improved |
| use channel-wise instead of ordinary temporal convolution in CSTM (ablation) | CSTM · Something-Something v1 val (8 frames) | top-1 accuracy: 46.9 -> 47.7 | Improved |
| increase input from 8 to 16 frames | STM · Something-Something v1 val | top-1 accuracy: 49.2 -> 50.7 | Improved |
| use STM instead of TSM at 8 frames vs 16 frames | STM 8F vs TSM 16F · Something-Something v1 val (8 frames) | top-1 accuracy: +2.7% accuracy | Improved |
Closest papers by meaning
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.95)
- StNet: Local and Global Spatial-Temporal Modeling for Action Recognition (AAAI 2019 · similarity 0.92)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.91)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.90)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 2019 · similarity 0.90)
- Spatio-Temporal Channel Correlation Networks for Action Classification (ECCV 2018 · 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.
