Research map / Video / Action recognition
PoTion: Pose MoTion Representation for Action Recognition
Vasileios Choutas, Philippe Weinzaepfel, Jérôme Revaud, Cordelia Schmid
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 PoTion to I3D | I3D† · JHMDB | mean per-class accuracy: 84.1 -> 85.5 | Improved |
| Add PoTion to I3D | I3D† · HMDB | mean per-class accuracy: 80.6 -> 80.9 | Improved |
| Add PoTion to I3D | I3D† · UCF101 | mean per-class accuracy: 97.7 -> 98.2 | Improved |
| Use intensity-only I aggregation instead of stacked U + I + N (ablation) | PoTion · JHMDB-1 | mean class accuracy: 58.5 ± 1.5 -> 52.2 ± 2.7 | Hurt |
| Use intensity-only I aggregation instead of stacked U + I + N (ablation) | PoTion · HMDB-1 | mean class accuracy: 44.4 ± 1.3 -> 43.3 ± 0.4 | No clear effect |
| Use U aggregation only instead of stacked U + I + N (ablation) | PoTion · HMDB-1 | mean class accuracy: 44.4 ± 1.3 -> 44.1 ± 0.9 | No clear effect |
| Use N aggregation only instead of stacked U + I + N (ablation) | PoTion · HMDB-1 | mean class accuracy: 44.4 ± 1.3 -> 42.5 ± 0.9 | Hurt |
| Add random horizontal flip augmentation (ablation) | PoTion training · JHMDB-1 | mean class accuracy: 51.3 ± 5.7 -> 58.5 ± 1.5 | Improved |
| Add random horizontal flip augmentation (ablation) | PoTion training · HMDB-1 | mean class accuracy: 43.4 ± 0.6 -> 44.4 ± 1.3 | No clear effect |
| Replace estimated pose with ground-truth puppet pose (ablation) | PoTion · JHMDB-1 | mean class accuracy: 58.5 ± 1.5 -> 62.1 ± 1.1 | Improved |
| Crop around the puppet when using ground-truth pose (ablation) | PoTion with puppet pose · JHMDB-1 | mean class accuracy: 62.1 ± 1.1 -> 67.9 ± 2.4 | Improved |
| Use 1 convolution per block instead of 2 (ablation) | PoTion classifier · JHMDB-1 | mean class accuracy: 58.5 ± 1.5 -> 54.1 ± 1.3 | Hurt |
Later papers that compared with this method
1 later accepted paper reports PoTion as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Hierarchical Self-Attention Network for Action Localization in Videos (ICCV 2019 · 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
- PA3D: Pose-Action 3D Machine for Video Recognition (CVPR 2019 · similarity 0.88)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.87)
- Multi-task Deep Learning for Real-Time 3D Human Pose Estimation and Action Recognition (TPAMI 2020 · similarity 0.87)
- MARS: Motion-Augmented RGB Stream for Action Recognition (CVPR 2019 · similarity 0.87)
- 2D/3D Pose Estimation and Action Recognition Using Multitask Deep Learning (CVPR 2018 · similarity 0.86)
- RPAN: An End-To-End Recurrent Pose-Attention Network for Action Recognition in Videos (ICCV 2017 · similarity 0.86)
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.
