Research map / Video / Action recognition
Fast Weakly Supervised Action Segmentation Using Mutual Consistency
Yaser Souri, Mohsen Fayyaz, Luca Minciullo, Gianpiero Francesca, Jüergen Gall
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 frame branch and mutual consistency loss (ablation) | segment generation branch · Breakfast split 1 | matching score: 0.729 -> 0.774 | Improved |
| use exponentially increasing dilation in temporal backbone instead of single conv (ablation) | segment branch · Breakfast split 1 | matching score: 0.691 -> 0.729 | Improved |
| use box mask instead of bell mask (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 43.6 -> 49.0 | Improved |
| add 10% overlap to box masks (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 49.0 -> 48.8 | No clear effect |
| add length regularizer and smoothing loss (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 47.4 -> 49.0 | Improved |
| add smoothing loss only (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 47.4 -> 47.5 | No clear effect |
| fuse both branches at inference (MuCon-full) instead of segment branch only (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 43.6 -> 49.0 | Improved |
| add teacher forcing in training (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 48.7 -> 49.0 | No clear effect |
| add MuCon loss in fully supervised training (ablation) | MuCon · Breakfast split 1 (fully supervised) | MoF accuracy: 63.7 -> 64.1 | Improved |
| add 10% fully annotated videos to weak supervision (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 49.0 -> 56.2 | Improved |
| add 1% fully annotated videos to weak supervision (ablation) | MuCon · Breakfast split 1 | MoF accuracy: 49.0 -> 51.0 | No clear effect |
| use MuCon instead of CDFL | weakly supervised action segmentation · Breakfast | training time (hours): 66.73 -> 4.57 | Improved |
Closest papers by meaning
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (CVPR 2024 · similarity 0.90)
- Weakly-Supervised Action Segmentation and Alignment via Transcript-Aware Union-of-Subspaces Learning (ICCV 2021 · similarity 0.90)
- Semi-Weakly-Supervised Learning of Complex Actions From Instructional Task Videos (CVPR 2022 · similarity 0.89)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.88)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
- Temporally Consistent Unbalanced Optimal Transport for Unsupervised Action Segmentation (CVPR 2024 · 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.
