Table 1:
The Dataset
| Name | Training instances | Held-out instances | Image size | The ground-truth factor |
|---|---|---|---|---|
| MNIST | 60, 000 | 10, 000 | 28 × 28 × 1 | discrete labels |
| Fashion-MNIST | 60, 000 | 10, 000 | 28 × 28 × 1 | discrete labels |
| Yale Faces | 1, 903 | 513 | 64 × 64 × 1 | azimuth lighting |
| 3D chairs | 69, 131 | 17, 237 | 64 × 64 × 3 | azimuth rotations |
| 3D cars | 14, 016 | 3, 552 | 64 × 64 × 3 | azimuth rotations |