TABLE 2:
Quantitative results for the fixed-domain UNIT setting on both facial and ImageNet images.
| Method | Facial Images | ImageNet Images | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||||
| seg→face | sketch→face | landmark→face | seg→cat | seg→dog | seg→plane | ||||||
| FID↓ | LPIPS↓ | FID↓ | FID↓ | LPIPS↓ | FID↓ | LPIPS↓ | FID↓ | LPIPS↓ | FID↓ | LPIPS↓ | |
| CycleGAN [1] | 132.6 | 0.550 | 120.8 | 328.1 | 0.607 | 278.3 | 0.709 | 317.1 | 0.745 | 172.1 | 0.716 |
| CUT [3] | 38.9 | 0.435 | 87.6 | 287.6 | 0.595 | 342.9 | 0.699 | 350.3 | 0.752 | 148.3 | 0.695 |
| MUNIT [24] | 212.1 | 0.623 | 266.4 | 184.0 | 0.665 | 339.9 | 0.820 | 372.4 | 0.902 | 312.7 | 0.764 |
|
| |||||||||||
| StarGANv2 [8] | 53.7 | 0.532 | 85.4 | 99.5 | 0.557 | - | - | - | - | - | - |
| DRIT++ [6] | 132.2 | 0.511 | 107.5 | 305.4 | 0.859 | - | - | - | - | - | - |
|
| |||||||||||
| Latent Anchor (ours) | 83.8 | 0.442 | 84.3 | 88.1 | 0.510 | 77.9 | 0.661 | 79.7 | 0.654 | 44.3 | 0.713 |