Class Incremental Learning on Tiny-ImageNet-200 T200-B100-S5 (test)
38.1Final AccuracyGIS (Ours)
Evaluation results
| Method | Links | |
|---|---|---|
| GIS (Ours)Backbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 38.1 | |
| IL2ABackbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 36.58 | |
| HFABackbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 36.11 | |
| EEILBackbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 35 | |
| iCaRLNMEBackbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 34.43 | |
| iCaRLCNNBackbone=ResNet-18, Batch size=64, Optimizer=Adam, Initial learning rate=0.001, Epochs per step=30, Memory buffer size=20 instances per class2023.04 | 23.17 |