Research map / Video / Action recognition
Spatiotemporal Pyramid Network 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 |
|---|---|---|---|
| Add the spatiotemporal pyramid network | BN-Inception two-stream model · UCF101 | accuracy: 94.0% -> 94.6% | Helped |
| Add the spatiotemporal pyramid network | BN-Inception two-stream model · HMDB51 | accuracy: 68.5% -> 68.9% | Helped |
| Use compact bilinear fusion instead of average fusion (ablation) | Two-stream BN-Inception · UCF101 (Split 1) | accuracy: 91.7% -> 93.2% | Helped |
| Add multi-path temporal fusion (ablation) | Spatiotemporal-fusion BN-Inception · UCF101 (Split 1) | accuracy: 93.2% -> 93.6% | Helped |
| Add spatiotemporal attention pooling (ablation) | Spatiotemporal and multi-temporal fusion model · UCF101 (Split 1) | accuracy: 93.6% -> 94.2% | Helped |
| Use compact bilinear fusion instead of average fusion (ablation) | BN-Inception fusion layer · UCF101 (Split 1) | accuracy: 91.7% -> 93.2% | Helped |
| Increase compact bilinear output dimension from 4096 to 8192 (ablation) | BN-Inception fusion layer · UCF101 (Split 1) | accuracy: 93.2% -> 93.2% | No clear effect |
| Reduce compact bilinear output dimension from 4096 to 2048 (ablation) | BN-Inception fusion layer · UCF101 (Split 1) | accuracy: 93.2% -> 92.9% | Hurt |
| Reduce compact bilinear output dimension from 4096 to 1024 (ablation) | BN-Inception fusion layer · UCF101 (Split 1) | accuracy: 93.2% -> 92.4% | Hurt |
| Fuse three temporal chunks with compact bilinear fusion instead of one chunk (ablation) | Optical-flow temporal stream · UCF101 | accuracy: 87.0% -> 89.3% | Helped |
| Increase temporal chunk count from three to five (ablation) | Optical-flow temporal stream with compact bilinear fusion · UCF101 | accuracy: 89.3% -> 89.2% | No clear effect |
| Use compact bilinear fusion for attention instead of average pooling (ablation) | Spatial BN-Inception stream · UCF101 | accuracy: 84.5% -> 86.6% | Helped |
Closest papers by meaning
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.92)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.92)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.92)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 2019 · similarity 0.90)
- Video Modeling With Correlation Networks (CVPR 2020 · similarity 0.90)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 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.