Research map / Video / Action recognition
Learn2Augment: Learning to Composite Videos for Data Augmentation in Action Recognition
Shreyank N Gowda, Marcus Rohrbach, Frank Keller, Laura Sevilla-Lara
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 the pairs Selector (ablation) | L2A · UCF101 (20% labeled) | accuracy: 55.8 -> 58.9 | Improved |
| add Video Compositing (ablation) | L2A · UCF101 (20% labeled) | accuracy: 54.5 -> 58.9 | Improved |
| add Semantic Matching (ablation) | L2A · UCF101 (20% labeled) | accuracy: 55.2 -> 58.9 | Improved |
| add all three components over the baseline (ablation) | L2A · UCF101 (20% labeled) | accuracy: 45.5 -> 58.9 | Improved |
| inpaint holes of removed objects in the background (ablation) | L2A · UCF101 (20% labeled) | accuracy: 57.6 -> 58.9 | Improved |
| use segmentation instead of bounding boxes (ablation) | L2A · UCF101 (20% labeled) | accuracy: 56.8 -> 58.9 | Improved |
| remove objects as well as humans from the foreground (ablation) | L2A · UCF101 (20% labeled) | accuracy: 55.7 -> 58.9 | Improved |
| mix semantically similar classes instead of the same class (ablation) | L2A · UCF101 (20% labeled) | accuracy: 41.4 -> 58.9 | Improved |
| re-train the classifier instead of joint training with the Selector (ablation) | L2A · UCF101 (20% labeled) | accuracy: 57.4 -> 60.3 | Improved |
| pre-train the Selector on Kinetics-400 and fix it (ablation) | L2A · UCF101 (20% labeled) | accuracy: 60.3 -> 64.8 | Improved |
| use the learned Selector instead of SMART-based selection (ablation) | L2A · UCF101 (20% labeled) | accuracy: 58.9 -> 60.3 | Improved |
| use semantic class matching instead of random matching in few-shot (ablation) | C3D-PN + L2A · HMDB51 few-shot (TruZe split, 5-shot) | accuracy: 47.7 -> 50.8 | Improved |
Closest papers by meaning
- Exploring Temporally Dynamic Data Augmentation for Video Recognition (ICLR 2023 · similarity 0.93)
- Markov Game Video Augmentation for Action Segmentation (ICCV 2023 · similarity 0.90)
- Selective Volume Mixup for Video Action Recognition (IJCV 2026 · similarity 0.89)
- Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition (CVPR 2019 · similarity 0.89)
- Two-Stage Active Learning for Efficient Temporal Action Segmentation (ECCV 2024 · similarity 0.88)
- 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.
