Domain Generalization on PACS, VLCS, OfficeHome, TerraIncognita, DomainNet (test)
96.1PACS AccuracySEDGE+
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SEDGE+Pre-training=CLIP, Ensemble=true2022.03 | 96.1 | 82.2 | 80.7 | 56.8 | 54.7 | 74.1 | |
| CADG+Pre-training=CLIP, Ensemble=true2022.03 | 94.6 | 82.2 | 79.5 | 55.7 | 51.6 | 72.7 | |
| EoA+Pre-training=CLIP, Ensemble=true2022.03 | 93.2 | 80.4 | 80.2 | 55.2 | 54.6 | 72.7 | |
| SPGBackbone=ResNet50, Protocol=Single-source DG2024.04 | 88.8 | 76.5 | 70.9 | 32.3 | 45.6 | 62.8 | |
| CoCoOpBackbone=ResNet50, Protocol=Single-source DG2024.04 | 88.1 | 68.2 | 70.6 | 25.6 | 45.5 | 59.6 | |
| DPLBackbone=ResNet50, Protocol=Single-source DG2024.04 | 86.8 | 75.2 | 70.8 | 28.4 | 43.8 | 61 | |
| CoOpBackbone=ResNet50, Protocol=Single-source DG2024.04 | 86 | 75.3 | 70.7 | 30.7 | 44.9 | 61.4 | |
| Lin. Prob.Backbone=ResNet50, Protocol=Single-source DG2024.04 | 77.3 | 65.5 | 46.4 | 23.3 | 6.3 | 43.8 | |
| random ensemblePre-training=CLIP, Ensemble=true2022.03 | 59.5 | 61.1 | 59.5 | 30.8 | 18.7 | 46 |