Research map / Video / Action recognition
A Generalized & Robust Framework for Timestamp Supervision in Temporal Action Segmentation
Rahul Rahaman, Dipika Singhania, Alexandre Thiery, Angela Yao
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 hard-boundary CE loss with E-M soft-weight loss (ablation) | EM-TSS · 50Salads (same timestamps as prior work) | MoF: 72.8 -> 77.4 | Improved |
| replace hard-boundary CE loss with E-M soft-weight loss (ablation) | EM-TSS · Breakfast (same timestamps as prior work) | F1@50: 36.7 -> 48.1 | Improved |
| add transition loss L_TR (ablation) | EM-TSS · 50Salads | Edit: 62.8 -> 66.2 | Improved |
| add transition loss L_TR (ablation) | EM-TSS · 50Salads | MoF: 77.4 -> 76.9 | No clear effect |
| add confidence loss L_Conf (ablation) | EM-TSS · 50Salads | Edit: 66.2 -> 71.6 | Improved |
| use informative length prior from one video instead of non-informative prior (ablation) | EM-TSS · 50Salads | MoF: 77.0 -> 77.3 | No clear effect |
| increase epochs per M-step N_max from 3 to 5 (ablation) | EM-TSS · 50Salads | MoF: 74.9 -> 77.9 | Improved |
| use start-frame timestamps instead of random (ablation) | EM-TSS · 50Salads | MoF: 77.3 -> 52.0 | Hurt |
| use E-M posterior instead of uniform mid-point boundaries | Uniform · 50Salads | MoF: 63.4 -> 77.3 | Improved |
| use E-M posterior instead of uniform boundary estimate | Uniform · 50Salads | boundary error (%): 1.23 -> 0.55 | Improved |
| use EM-Gen instead of EM-TSS with 20% missed segments (ablation) | EM-TSS · 50Salads (20% missing) | MoF: 65.6 -> 71.9 | Improved |
| use EM-Gen instead of Naive under SkipTag | Naive · 50Salads (SkipTag) | MoF: 70.4 -> 74.4 | Improved |
Closest papers by meaning
- Temporal Action Segmentation From Timestamp Supervision (CVPR 2021 · similarity 0.94)
- Reducing the Label Bias for Timestamp Supervised Temporal Action Segmentation (CVPR 2023 · similarity 0.90)
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (CVPR 2024 · similarity 0.86)
- OnlineTAS: An Online Baseline for Temporal Action Segmentation (NeurIPS 2024 · similarity 0.86)
- HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · similarity 0.86)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · 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.
