Table 3 The results of evaluating ResNet-50 on the PACS dataset by excluding one domain as the target domain.
From: ICRL: independent causality representation learning for domain generalization
Methods | Art | Cartoon | Photo | Sketch | Avg. |
|---|---|---|---|---|---|
DeepAll63 | 84.94 | 76.98 | 97.64 | 76.75 | 84.08 |
MetaReg73 | 87.20 | 79.20 | 97.60 | 70.30 | 83.60 |
MASF75 | 82.89 | 80.49 | 95.01 | 72.29 | 82.67 |
EISNet75 | 86.64 | 81.53 | 97.11 | 78.07 | 85.84 |
MatchDG76 | 85.61 | 82.12 | 97.94 | 78.76 | 86.11 |
FACT55 | 90.89 | 83.65 | 97.78 | 86.17 | 89.62 |
CIRL47 | 90.67 | 84.04 | 97.12 | 86.17 | 89.50 |
ICRL (WGAN) | 90.83 | 84.55 | 97.60 | 87.90 | 90.22 |
ICRL (Vanilla GAN) | 86.62 | 79.44 | 96.65 | 86.56 | 87.31 |
ICRL (WGAN-GP) | 89.84 | 84.12 | 97.06 | 86.97 | 89.49 |
