Research map / Video / Action recognition
Pre-training for Action Recognition with Automatically Generated Fractal Datasets
Davyd Svyezhentsev, George Retsinas, Petros A. Maragos
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 |
|---|---|---|---|
| use supervised classification pre-training instead of MoCoV2 (ablation) | fractal video pre-training · HMDB51 | top-1 accuracy: 54.3 -> 61.5 | Improved |
| use supervised classification pre-training instead of MoCoV2 (ablation) | fractal video pre-training · UCF101 | top-1 accuracy: 80.2 -> 84.9 | Improved |
| use SimCLR instead of MoCoV2 (ablation) | fractal video pre-training · HMDB51 | top-1 accuracy: 54.3 -> 57.5 | Improved |
| use BYOL instead of MoCoV2 (ablation) | fractal video pre-training · HMDB51 | top-1 accuracy: 54.3 -> 52.0 | Hurt |
| use fractal videos instead of octopus videos (ablation) | MoCoV2 pre-training · HMDB51 | top-1 accuracy: 50.9 -> 54.3 | Improved |
| use fractal videos instead of dead leaves (ablation) | MoCoV2 pre-training · HMDB51 | top-1 accuracy: 39.9 -> 54.3 | Improved |
| use fractal videos instead of Perlin noise (ablation) | MoCoV2 pre-training · UCF101 | top-1 accuracy: 72.5 -> 80.2 | Improved |
| add amplified diversity (nonlinear variations) (ablation) | MoCoV2 fractal pre-training · HMDB51 | top-1 accuracy: 47.3 -> 50.4 | Improved |
| add background randomization (ablation) | MoCoV2 fractal pre-training · DIVING48 | top-1 accuracy: 24.3 -> 23.0 | Hurt |
| add background randomization (ablation) | MoCoV2 fractal pre-training · HMDB51 | top-1 accuracy: 41.4 -> 47.4 | Improved |
| add random perspective (ablation) | MoCoV2 fractal pre-training · HMDB51 | top-1 accuracy: 54.3 -> 53.3 | Hurt |
| add perspective augmentation (ablation) | supervised fractal pre-training (1000 classes) · HMDB51 | top-1 accuracy: 62.4 -> 65.4 | Improved |
Closest papers by meaning
- Generating Human Action Videos by Coupling 3D Game Engines and Probabilistic Graphical Models (IJCV 2019 · similarity 0.89)
- Procedural Generation of Videos to Train Deep Action Recognition Networks (CVPR 2017 · similarity 0.88)
- Synthetic Humans for Action Recognition from Unseen Viewpoints (IJCV 2021 · similarity 0.88)
- How Transferable are Video Representations Based on Synthetic Data? (NeurIPS 2022 · similarity 0.87)
- Taylor Videos for Action Recognition (ICML 2024 · similarity 0.85)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 2024 · similarity 0.85)
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.
