Class-incremental image classification on NICO++ C (test)
90.38Accuracy (T1)FedET
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FedETBackbone=ResNet-182026.02 | 90.38 | 52.85 | 41.71 | 33.24 | 37.98 | 31.62 | |
| FedIL+Backbone=ResNet-182026.02 | 90.16 | 56.33 | 49.77 | 46.82 | 45.26 | 41.88 | |
| FedAvgBackbone=ResNet-182026.02 | 89.03 | 51.68 | 40.75 | 32.19 | 38.01 | 30.56 | |
| FedProxBackbone=ResNet-182026.02 | 88.96 | 51.61 | 40.7 | 32.19 | 37.99 | 30.58 | |
| FedEWCBackbone=ResNet-182026.02 | 88.96 | 51.85 | 40.68 | 32.33 | 37.82 | 30.52 | |
| OSCAR-ILBackbone=ResNet-182026.02 | 79.67 | 35.45 | 23.97 | 28.02 | 23.84 | 22.45 | |
| OSCAR-RBackbone=ResNet-182026.02 | 79.67 | 35.67 | 23.57 | 28.51 | 23.9 | 22.41 | |
| OSI-FLBackbone=ResNet-18, Retained samples p=52026.02 | 79.67 | 60.39 | 59.43 | 54.96 | 48.25 | 49.76 |