Research map / Video / Action recognition
Spatiotemporal Joint Filter Decomposition in 3D Convolutional Neural Networks
Zichen Miao, Ze Wang, Xiuyuan Cheng, Qiang Qiu
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 |
|---|---|---|---|
| dilate spatial atoms by 2 at test when training on downsampled spatial data (ablation) | STDCF · KTH | accuracy: 53.08 -> 75.34 | Improved |
| dilate temporal atoms by 2 when learning on downsampled temporal data (ablation) | STDCF · KTH | accuracy: 66.49 -> 77.11 | Improved |
| replace regular 3D filters with STDCF (full-size learning) | 3-layer 3D CNN · KTH | accuracy: 81.26 -> 85.61 | Improved |
| replace Rank-1 3D with STDCF (full-size learning) | 3-layer 3D CNN · KTH | accuracy: 82.55 -> 85.61 | Improved |
| use iterative two-stage learning (ITSL) instead of full-size learning | Reg. 3D · KTH | accuracy: 81.26 -> 76.94 | Hurt |
| use ITSL instead of full-size learning | Rank-1 3D · KTH | accuracy: 82.55 -> 79.10 | Hurt |
| use ITSL instead of full-size learning | STDCF · KTH | accuracy: 85.61 -> 84.93 | No clear effect |
| use ITSL instead of full-size learning | STDCF · KTH | max GPU memory: 459.4M -> 228.9M | Improved |
| add iterations 2 and 3 of two-stage learning (ablation) | STDCF · KTH | accuracy: 75.34 -> 84.93 | Improved |
| add iterations 2 and 3 of two-stage learning (ablation) | Reg. 3D · KTH | accuracy: 74.85 -> 76.94 | No clear effect |
| add iterations 2 and 3 of two-stage learning (ablation) | Rank-1 3D · KTH | accuracy: 73.85 -> 79.1 | Mixed |
| stage-s update (ITSL) over stage-t model, E=3 (ablation) | STDCF-R50 · Kinetics-400 | top-1 accuracy: 73.6 -> 74.0 | Improved |
Closest papers by meaning
- Grouped Spatial-Temporal Aggregation for Efficient Action Recognition (ICCV 2019 · similarity 0.88)
- Learning Spatio-Temporal Representation With Pseudo-3D Residual Networks (ICCV 2017 · similarity 0.88)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.88)
- In Defense of Image Pre-training for Spatiotemporal Recognition (ECCV 2022 · similarity 0.87)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.87)
- Spatio-Temporal Channel Correlation Networks for Action Classification (ECCV 2018 · similarity 0.87)
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.
