Image Generation on CelebA 64x64 (train)
85.83PrecisionDenseFlow
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DenseFlow2023.05 | 85.83 | — | 38.22 | 117 | 82 | |
| BigGANvariant=hard truncation, psi=12023.05 | 84.3 | 17.97 | 37.46 | 111 | 49 | |
| BigGAN (PR-divergence)lambda=52023.05 | 83.96 | 22.45 | 25.81 | 121 | 43 | |
| BigGANvariant=hard truncation, psi=0.52023.05 | 83.7 | 25.7 | 28.81 | 108 | 42 | |
| BigGAN (PR-divergence)lambda=0.22023.05 | 83.37 | 8.79 | 44.07 | 109 | 54 | |
| BigGAN (PR-divergence)lambda=22023.05 | 82.98 | 14.23 | 32.87 | 116 | 49 | |
| BigGAN (PR-divergence)lambda=12023.05 | 81.7 | 13.07 | 36.85 | 100 | 47 | |
| BigGANvariant=hard truncation, psi=22023.05 | 80.81 | 10.6 | 48.21 | 96 | 50 | |
| BigGANvariant=baseline (hinge loss)2023.05 | 78.41 | 9.16 | 51.12 | 89 | 48 | |
| BigGAN (PR-divergence)lambda=0.52023.05 | 77.6 | 6.03 | 55.98 | 88 | 50 | |
| ADM-IP2023.05 | 23.42 | 1.53 | 64.48 | 9 | 24 |