Research map / Video / Action recognition
Weakly-Supervised Action Segmentation and Alignment via Transcript-Aware Union-of-Subspaces Learning
Zijia Lu, Ehsan Elhamifar
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 subspace assignment score sigma to the loss (ablation) | TASL(3) · Breakfast segmentation | IoU: 11.0 -> 35.2 | Improved |
| use discriminative distance score psi instead of plain distance (ablation) | TASL(3) · Breakfast segmentation | IoU: 15.5 -> 35.2 | Improved |
| add positive and negative soft alignments (ablation) | TASL(3) · Breakfast segmentation | IoU: 27.1 -> 35.2 | Improved |
| add the L_reg regularization to the Viterbi alignment (ablation) | TASL(3) · Breakfast segmentation | IoU: 3.5 -> 35.2 | Improved |
| use TASL instead of CDFL | TASL(3) · CrossTask segmentation | Mof: 31.9 -> 40.7 | Improved |
| use TASL instead of CDFL | TASL(10,3) · Hollywood segmentation | Mof: 39.9 -> 43.7 | Improved |
| use TASL instead of CDFL (best run) | TASL(3) · Breakfast segmentation | Mof: 50.8 -> 49.9 | Hurt |
| use hierarchical segmentation with representative transcripts (ablation) | TASL(3) · Breakfast | inference time: 1.3s -> 0.3s | Improved |
| use only representative transcripts (Rep) (ablation) | TASL(3) · Breakfast | IoU: 35.2 -> 35.0 | No clear effect |
| increase background subspace dimension to 10 (ablation) | TASL(10,3) · Hollywood Extended | IoD-bg: 17.4 -> 19.2 | Improved |
| vary subspace dimension of all actions (3, 5, 10/3) (ablation) | TASL · Breakfast segmentation | IoU: 35.2 -> 35.1 | No clear effect |
Closest papers by meaning
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (CVPR 2024 · similarity 0.92)
- Semi-Weakly-Supervised Learning of Complex Actions From Instructional Task Videos (CVPR 2022 · similarity 0.91)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · similarity 0.90)
- Fast Weakly Supervised Action Segmentation Using Mutual Consistency (TPAMI 2021 · similarity 0.90)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.89)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.89)
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.
