Research map / Video / Action recognition
Spatio-Temporal Channel Correlation Networks for Action Classification
Ali Diba, Mohsen Fayyaz, Vivek Sharma, M. Mahdi Arzani, Rahman Yousefzadeh, Juergen Gall, 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 STC blocks to 3D-ResNet101 (training from scratch) | 3D-ResNet 101 · UCF101 split 1 (from scratch) | accuracy: 46.7 -> 47.9 | Improved |
| use both TCB and SCB branches instead of SCB only (ablation) | STC block · UCF101 split 1 (from scratch) | accuracy: 46.1 -> 47.9 | Improved |
| use both TCB and SCB branches instead of TCB only (ablation) | STC block · UCF101 split 1 (from scratch) | accuracy: 47.2 -> 47.9 | Improved |
| increase network depth from 18 to 101 (ablation) | STC-ResNet · UCF101 split 1 (from scratch) | accuracy: 42.8 -> 47.9 | Improved |
| increase input temporal depth from 16 to 32 frames (ablation) | STC-ResNet101 · UCF101 split 1 (from scratch) | accuracy: 45.6 -> 47.9 | Improved |
| change frame sampling stride from 1 to 2 (ablation) | STC-ResNet101 · UCF101 split 1 | accuracy: 44.6% -> 47.9% | Improved |
| add 2D-to-3D knowledge transfer initialization to STC-ResNext (fine-tune on UCF101) (ablation) | STC-ResNext · UCF101 split 1 (transfer) | accuracy: 84.7 -> 88.2 | Improved |
| add STC blocks to 3D-ResNext101 on Kinetics | 3D ResNext101 · Kinetics val | top-1 accuracy: 65.1 -> 66.2 | Improved |
| add STC blocks to 3D-ResNet101 on UCF101 (Kinetics pretrained) | 3D ResNet 101 · UCF101 (3 splits avg) | accuracy: 88.9 -> 90.1 | Improved |
| add 2D-to-3D transferred knowledge initialization to STC-ResNet101 | STC-ResNet 101 · UCF101 (3 splits avg) | accuracy: 90.1 -> 92.6 | Improved |
| add STC blocks to 3D-ResNext101 on HMDB51 | 3D ResNext 101 · HMDB51 (3 splits avg) | accuracy: 63.8 -> 65.4 | Improved |
Closest papers by meaning
- Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? (CVPR 2018 · similarity 0.91)
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.91)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.91)
- StNet: Local and Global Spatial-Temporal Modeling for Action Recognition (AAAI 2019 · similarity 0.90)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.90)
- STM: SpatioTemporal and Motion Encoding for Action Recognition (ICCV 2019 · 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.
