notesum.ai
Published at November 26Pre-training for Action Recognition with Automatically Generated Fractal Datasets
cs.CV
Released Date: November 26, 2024
Authors: Davyd Svyezhentsev, George Retsinas, Petros Maragos

| Method | Kept | HMDB51 | UCF101 | DIVING48 | EGTEA | VOLLEY | YUP |
|---|---|---|---|---|---|---|---|
| Scratch | - | 31.5 | 70.3 | 4.7 | 50.5 | 53.8 | 43.7 |
| Initial | - | 41.4 | 72.7 | 24.3 | 49.8 | 80.6 | 52.3 |
| Background | Yes | 47.4 | 74.2 | 23.0 | 47.1 | 77.6 | 56.4 |
| Motion | Yes | 47.3 | 75.7 | 21.7 | 48.6 | 79.2 | 59.5 |
| Diversity | Yes | 50.4 | 77.8 | 24.9 | 49.8 | 80.8 | 63.1 |
| Scale + Shake | Yes | 52.8 | 78.0 | 26.0 | 50.7 | 80.8 | 61.9 |
| Shift | Yes | 54.5 | 79.6 | 29.5 | 50.8 | 80.9 | 65.1 |
| Zoom | Yes | 54.3 | 80.2 | 30.8 | 52.0 | 81.4 | 65.2 |
| Perspective | No | 53.3 | 78.7 | 25.0 | 50.0 | 80.7 | 62.6 |
| Group | No | 53.3 | 79.7 | 30.3 | 51.9 | 81.9 | 66.0 |