Research map / Video / Action recognition
Spatiotemporal Multiplier Networks for Video Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace additive motion residual fusion with multiplicative motion gating in appearance residual units (ablation) | Two-stream ResNet · UCF101 split 1 | classification error: 9.38 -> 8.72 | Helped |
| Replace additive motion residual fusion with multiplicative motion gating in appearance residual units (ablation) | Two-stream ResNet · HMDB51 split 1 | classification error: 41.89 -> 37.23 | Helped |
| Add direct additive cross-stream connections from motion to appearance (ablation) | Two-stream ResNet · UCF101 split 1 | classification error: 24.78 | Hurt |
| Add direct multiplicative cross-stream connections from motion to appearance (ablation) | Two-stream ResNet · HMDB51 split 1 | classification error: 77.89 | Hurt |
| Reverse additive residual fusion so appearance feeds motion (ablation) | Two-stream ResNet · UCF101 split 1 | classification error: 16.76 | Hurt |
| Use bidirectional multiplicative residual gating (ablation) | Two-stream ResNet · HMDB51 split 1 | classification error: 48.56 | Hurt |
| Add identity-initialized temporal filters (ablation) | Multiplicative-gating network · UCF101 split 1 | classification error: 8.72 -> 7.45 | Helped |
| Add identity-initialized temporal filters (ablation) | Multiplicative-gating network · HMDB51 split 1 | classification error: 37.23 -> 34.38 | Helped |
| Use temporal max-pooling with center-initialized filters in every residual unit (ablation) | Multiplicative-gating network · UCF101 split 1 | classification error: 7.79 -> 6.00 | Helped |
| Use temporal max-pooling with center-initialized filters in every residual unit (ablation) | Multiplicative-gating network · HMDB51 split 1 | classification error: 34.38 -> 30.98 | Helped |
| Fuse the proposed ConvNet with IDT features | Ours · UCF101 | accuracy: 94.2% -> 94.9% | Helped |
| Fuse the proposed ConvNet with IDT features | Ours · HMDB51 | accuracy: 68.9% -> 72.2% | Helped |
Closest papers by meaning
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.90)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.90)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.88)
- Recognize Actions by Disentangling Components of Dynamics (CVPR 2018 · similarity 0.88)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.88)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · similarity 0.88)
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.