Research map / Video / Action recognition
Multi-Modal Few-Shot Temporal Action Segmentation
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 typed relational edges to the Action Relation Graph (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 8.9 -> 15.0 | Improved |
| use the dense graph instead of a sparse hierarchical graph (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 10.3 -> 15.0 | Improved |
| add task nodes to the graph (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 8.6 -> 15.0 | Improved |
| learn dynamic edge weights alpha instead of fixed alpha (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 12.1 -> 15.0 | Improved |
| use per-head alpha instead of shared alpha across heads (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 11.9 -> 15.0 | Improved |
| add prediction fusion between modalities (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 12.7 -> 15.0 | Improved |
| fuse predictions in the first SHCA layer instead of the last (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 13.5 -> 15.0 | Improved |
| add L2 penalty on influence factor tau_1 (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 11.1 -> 15.0 | Improved |
| use all three refinement stages instead of the first stage (ablation) | PGNet (COIN, 3way-3shot) · COIN novel tasks | F1@50: 11.4 -> 15.0 | Improved |
| add unlabeled support videos to zero-shot textual PGNet | PGNet · COIN novel tasks | F1@50: 8.6 -> 10.2 | Improved |
| add weak labels (one frame per action instance) to unlabeled few-shot | PGNet · COIN novel tasks | F1@50: 10.2 -> 13.8 | Improved |
| train on COIN instead of CrossTask base tasks | PGNet (5way-3shot) · CrossTask novel tasks | Edit and F1@50: Edit 34.7 -> 41.5; F1@50 17.5 -> 12.4 | Mixed |
Closest papers by meaning
- Understanding Multi-Task Activities from Single-Task Videos (CVPR 2025 · similarity 0.90)
- Temporal Action Segmentation: An Analysis of Modern Techniques (TPAMI 2023 · similarity 0.90)
- Two-Stage Active Learning for Efficient Temporal Action Segmentation (ECCV 2024 · similarity 0.89)
- 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 0.88)
- Condensing Action Segmentation Datasets via Generative Network Inversion (CVPR 2025 · 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.
