Research map / Video / Action recognition
Markov Game Video Augmentation for Action Segmentation
Nicolas Aziere, Sinisa Todorovic
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 MVGA data augmentation | MS-TCN · Breakfast | MoF: 66.3 -> 71.1 | Improved |
| add MVGA data augmentation | ASFormer · 50Salads | MoF: 85.6 -> 86.0 | Improved |
| add MVGA data augmentation | MS-TCN · GTEA | MoF: 76.3 -> 79.6 | Improved |
| use RL-trained Actor-Critic for frame-feature augmentation instead of Gaussian noise (ablation) | MS-TCN · GTEA | MoF: 77.5 -> 79.6 | Improved |
| use RL instead of supervised (Non-RL) meta-model for feature augmentation (ablation) | MS-TCN · GTEA | MoF: 77.7 -> 79.6 | Improved |
| add second agent (frame selection policy mu2) to single-agent feature augmentation (ablation) | MVGA · GTEA | MoF: 78.2 -> 79.6 | Improved |
| triple the training set with 20% new transcripts (ablation) | MS-TCN + MVGA · GTEA | MoF: 78.0 -> 79.6 | Improved |
| augment with too many (5x) generated videos (ablation) | MS-TCN + MVGA · GTEA | MoF: 79.6 -> 78.2 | Hurt |
| add new transcripts/videos without frame-feature augmentation (ablation) | MS-TCN · GTEA | MoF: 78.0 -> 74.9 | Hurt |
| generate transcripts with UCT instead of auto-regressive Transformer (ablation) | transcript generation · GTEA | edit distance to ground truth (top 10%): 19.5 -> 14.0 | Improved |
| use MVGA instead of Speed or C2F augmentation | MS-TCN · Breakfast | MoF: 66.5 -> 71.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MSTCN — MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation (CVPR 2019) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learn2Augment: Learning to Composite Videos for Data Augmentation in Action Recognition (ECCV 2022 · similarity 0.90)
- Data Augmentation for Instruction Following Policies via Trajectory Segmentation (AAAI 2025 · similarity 0.89)
- Weakly-Supervised Action Segmentation and Alignment via Transcript-Aware Union-of-Subspaces Learning (ICCV 2021 · similarity 0.89)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
- Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment (CVPR 2024 · similarity 0.87)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 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.
