Research map / Video / Action recognition
Semantic Image Networks for Human Action Recognition
Sunder Ali Khowaja, Seok-Lyong Lee
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 |
|---|---|---|---|
| replace single dynamic image with single SemI (segmentation plus background overlay) (ablation) | ResNeXt101 · UCF101 split 1 | mean class accuracy: 71.6 % -> 73.1% | Improved |
| replace Single dynamic map with single SemM | ResNeXt101 · UCF101 | mean class accuracy: 69.2% -> 72.5% | Improved |
| replace single dynamic map with single SemM | ResNeXt101 · HMDB51 | mean class accuracy: 39.8% -> 42.2% | Improved |
| move rank pooling to a deeper branching point (RIPB1 to RIPB2) (ablation) | Inception-ResNetv2 SemM · UCF101 | mean class accuracy: 73.8% -> 75.9% | Improved |
| swap ResNeXt101 for Inception-ResNetv2 backbone | single SemI · UCF101 split 1 | mean class accuracy: 73.1% -> 74.7% | Improved |
| increase window from 15 to 20 (stride 6) (ablation) | multiple SemI Inception-ResNetv2 · UCF101 | mean class accuracy: 88.7% -> 80.6% | Hurt |
| increase stride from 6 to 9 at window 15 (ablation) | multiple SemI · UCF101 | mean class accuracy: 88.2% -> 88.0% | No clear effect |
| replace multiple dynamic images with multiple SemI | Inception-ResNetv2 · HMDB51 | mean class accuracy: 58.2% -> 59.9% | Improved |
| replace RGB static images with multiple SemI | Inception-ResNetv2 · UCF101 | mean class accuracy: 88.6% -> 88.9% | No clear effect |
| replace RGB static images with multiple SemI | Inception-ResNetv2 · HMDB51 | mean class accuracy: 54.6% -> 59.9% | Improved |
| replace warped optical flows with SemOF | ResNeXt101 · HMDB51 | mean class accuracy: 56.0% -> 59.3% | Improved |
| replace dynamic optical flows with SemOF | ResNeXt101 · UCF101 | mean class accuracy: 86.9% -> 87.1% | No clear effect |
Closest papers by meaning
- Unsupervised Deep Learning of Mid-Level Video Representation for Action Recognition (AAAI 2018 · similarity 0.89)
- Temporal Segment Networks for Action Recognition in Videos (TPAMI 2018 · similarity 0.89)
- Lattice Long Short-Term Memory for Human Action Recognition (ICCV 2017 · similarity 0.89)
- Harnessing Object and Scene Semantics for Large-Scale Video Understanding (CVPR 2016 · similarity 0.88)
- Temporal Segment Networks: Towards Good Practices for Deep Action Recognition (ECCV 2016 · similarity 0.88)
- Action Recognition with Dynamic Image Networks (TPAMI 2017 · 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.
