Research map / Video / Action recognition
Towards Generalizing Temporal Action Segmentation to Unseen Views
Emad Bahrami, Olga Zatsarynna, Gianpiero Francesca, 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 sequence loss and action loss together (ablation) | Baseline TAS model (MViTv2) · Assembly101 unseen ego views | F1@50: 6.3 -> 9.7 | Improved |
| add sequence loss and action loss together (ablation) | Baseline TAS model (MViTv2) · Assembly101 unseen exo views | F1@50: 11.7 -> 13.2 | Improved |
| add sequence loss and action loss together (ablation) | Baseline TAS model (MViTv2) · Assembly101 seen views | F1@50: 31.1 -> 32.1 | Improved |
| add sequence loss alone (ablation) | Baseline TAS model (MViTv2) · Assembly101 unseen ego views | F1@50: 6.3 -> 8.1 | Improved |
| add action loss alone (ablation) | Baseline TAS model (MViTv2) · Assembly101 unseen ego views | F1@50: 6.3 -> 8.5 | Improved |
| include same-view action segments in action loss (ablation) | Ours · Assembly101 unseen ego views | F1@50: 9.7 -> 7.0 | Hurt |
| add adversarial view-classification loss (ablation) | Ours · Assembly101 unseen ego views | F1@50: 9.7 -> 6.4 | Hurt |
| use cosine similarity instead of MSE (ablation) | Ours · Assembly101 unseen ego views | F1@50: 7.2 -> 9.7 | Improved |
| use cosine similarity instead of KL divergence (ablation) | Ours · Assembly101 unseen ego views | F1@50: 6.5 -> 9.7 | Improved |
| use fine instead of coarse action segments in action loss (ablation) | Ours · Assembly101 unseen ego views | F1@50: 6.5 -> 9.7 | Improved |
| desynchronize views by random shift of 1 frame (ablation) | Ours · Assembly101 unseen ego views | F1@50: 9.7 -> 7.7 | Hurt |
| desynchronize views by random shift of 10 frames (ablation) | Ours · Assembly101 unseen ego views | F1@50: 9.7 -> 6.3 | Hurt |
Closest papers by meaning
- OnlineTAS: An Online Baseline for Temporal Action Segmentation (NeurIPS 2024 · similarity 0.89)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.89)
- Progress-Aware Online Action Segmentation for Egocentric Procedural Task Videos (CVPR 2024 · similarity 0.89)
- ActFusion: a Unified Diffusion Model for Action Segmentation and Anticipation (NeurIPS 2024 · similarity 0.89)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.89)
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (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.
