Research map / Video / Action recognition
A Hybrid RNN-HMM Approach for Weakly Supervised Temporal Action Segmentation
Hilde Kuehne, Alexander Richard, Jüergen Gall
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 subaction modeling (split actions into subactions) (ablation) | GRU + HMM · Breakfast (weak) | MoF: 22.4 -> 28.8 | Improved |
| add subaction number reestimation (ablation) | GRU + subactions · Breakfast (weak) | MoF: 28.8 -> 33.3 | Improved |
| set subaction number from ground-truth action lengths (oracle) (ablation) | GRU + subactions · Breakfast (weak) | MoF: 33.3 -> 51.3 | Improved |
| use GRU instead of MLP for frame classifier (ablation) | HMM hybrid · Breakfast (weak) | MoF: 23.1 -> 29.3 (iteration 4) | Improved |
| use GRU instead of GMM (ablation) | HMM hybrid · Breakfast (weak) | MoF: 27.0 -> 29.3 (iteration 4) | Improved |
| add length prior (linear decay) (ablation) | GRU + reestimation · Breakfast (weak) | MoF: 32.6 -> 37.0 | Improved |
| add length prior (half Gaussian) (ablation) | GRU + reestimation · Breakfast (weak) | MoF: 32.6 -> 36.7 | Improved |
| add length prior (box function) (ablation) | GRU + reestimation · Hollywood Extended (weak) | IoU: 11.5 -> 9.9 | Hurt |
| add length prior (half Gaussian) (ablation) | GRU + reestimation · Hollywood Extended (weak) | IoU: 11.5 -> 12.3 | Improved |
| add length prior (ablation) | GRU + reestimation · Hollywood Extended (alignment) | Jaccard IoD: 46.3 -> 46.0 | No clear effect |
| add length prior (ablation) | GRU · Breakfast (fully supervised) | MoF: 60.2 -> 61.3 | Improved |
| add length prior (reduces skip states) (ablation) | GRU + reestimation · Breakfast (weak) | fraction of skip states: 74.6% -> 61.9% | Improved |
Closest papers by meaning
- Weakly Supervised Action Learning With RNN Based Fine-To-Coarse Modeling (CVPR 2017 · similarity 0.95)
- Weakly-Supervised Action Segmentation With Iterative Soft Boundary Assignment (CVPR 2018 · similarity 0.92)
- Weakly Supervised Energy-Based Learning for Action Segmentation (ICCV 2019 · similarity 0.91)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · similarity 0.91)
- SCT: Set Constrained Temporal Transformer for Set Supervised Action Segmentation (CVPR 2020 · similarity 0.90)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · 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.
