Research map / Video / Action recognition
Hierarchical Self-Attention Network for Action Localization in Videos
Rizard Renanda Adhi Pramono, Yie-Tarng Chen, Wen-Hsien Fang
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 HISAN hierarchical self-attention (ablation) | two-stream Faster R-CNN · UCF101-24 | video mAP@0.5: 43.69 -> 47.28 | Improved |
| add HISAN hierarchical self-attention (ablation) | two-stream Faster R-CNN · UCF101-24 | frame mAP@0.5: 63.80 -> 70.47 | Improved |
| add HISAN hierarchical self-attention (ablation) | two-stream Faster R-CNN · J-HMDB | video mAP@0.5: 71.24 -> 82.98 | Improved |
| add HISAN hierarchical self-attention (ablation) | two-stream Faster R-CNN · J-HMDB | frame mAP@0.5: 60.21 -> 76.04 | Improved |
| add sequence rescoring (SR) (ablation) | HISAN · UCF101-24 | video mAP@0.5: 47.28 -> 48.44 | Improved |
| add sequence rescoring (SR) (ablation) | HISAN · J-HMDB | video mAP@0.5: 82.98 -> 83.27 | Improved |
| add motion saliency fusion (ablation) | HISAN + SR · UCF101-24 | frame mAP@0.5: 71.03 -> 73.71 | Improved |
| add motion saliency fusion (ablation) | HISAN + SR · UCF101-24 | video mAP@0.5: 48.44 -> 49.50 | Improved |
| add motion saliency fusion (ablation) | HISAN + SR · J-HMDB | frame mAP@0.5: 76.53 -> 76.72 | Improved |
| replace VGG-16 with ResNet-101+FPN backbone | HISAN full model · UCF101-24 | video mAP@0.5: 49.50 -> 51.47 | Improved |
| replace VGG-16 with ResNet-101+FPN backbone | HISAN full model · J-HMDB | video mAP@0.5: 84.02 -> 86.49 | Improved |
| compare capsule network (Duarte et al.) at stricter IoU than 0.2 | Duarte et al. · J-HMDB | video mAP: 95.40 -> 61.95 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PoTion — PoTion: Pose MoTion Representation for Action Recognition (CVPR 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Class Semantics-Based Attention for Action Detection (ICCV 2021 · similarity 0.88)
- Weakly-Supervised Action Localization, and Action Recognition Using Global–Local Attention of 3D CNN (IJCV 2022 · similarity 0.88)
- Temporal Action Localization in Untrimmed Videos via Multi-Stage CNNs (CVPR 2016 · similarity 0.87)
- SSAN: Separable Self-Attention Network for Video Representation Learning (CVPR 2021 · similarity 0.87)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.87)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · 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.
