Research map / Video / Action recognition
MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a single-stage network with four sequential refinement stages (ablation) | TCN for temporal action segmentation · 50Salads | F1@10: 27.0 -> 76.3 | Helped |
| Use four-stage refinement rather than a parameter-matched 48-layer single-stage network (ablation) | TCN for temporal action segmentation · 50Salads | F1@10: 49.0 -> 76.3 | Helped |
| Add truncated MSE smoothing loss to cross-entropy training (ablation) | MS-TCN · 50Salads | F1@10: 71.3 -> 76.3 | Helped |
| Replace truncated MSE smoothing loss with KL loss (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 71.9 | Hurt |
| Increase the smoothing-loss weight from lambda=0.15 to lambda=0.25 (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 74.7 | Hurt |
| Increase the smoothing-loss truncation threshold from tau=4 to tau=5 (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 66.6 | Hurt |
| Add convolutional features to the probabilities passed to higher stages (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 56.2 | Hurt |
| Reduce input temporal resolution from 15 fps to 1 fps (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 77.8 | Helped |
| Increase the number of dilated layers per stage from 10 to 12 (ablation) | MS-TCN · 50Salads | F1@10: 76.3 -> 77.8 | Helped |
| Fine-tune the I3D features (ablation) | MS-TCN · GTEA | F1@10: 85.8 -> 87.5 | Helped |
| Use IDT instead of I3D features (ablation) | MS-TCN · Breakfast | F1@10: 52.6 -> 58.2 | Helped |
Closest papers by meaning
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.95)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.91)
- MS-TCT: Multi-Scale Temporal ConvTransformer for Action Detection (CVPR 2022 · similarity 0.90)
- Weakly-Supervised Action Segmentation With Iterative Soft Boundary Assignment (CVPR 2018 · similarity 0.90)
- Iterative Contrast-Classify for Semi-supervised Temporal Action Segmentation (AAAI 2022 · similarity 0.90)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (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.