Class Incremental Learning on CIFAR100 20 steps B0 (test)
80.41AccuracyJoint Training (Bound)
Evaluation results
| Method | Links | |
|---|---|---|
| Joint Training (Bound)2022.04 | 80.41 | |
| DER (without pruning)Backbone=modified 32-layer ResNet, #Paras=4.83M2021.03 | 71.07 | |
| FOSTERBackbone models=Single backbone2022.04 | 70.65 | |
| DERBackbone=modified 32-layer ResNet, #Paras=0.45M2021.03 | 68.82 | |
| DERBackbone models=Same number as incremental sessions2022.04 | 67.98 | |
| WA2022.04 | 64.64 | |
| WABackbone=modified 32-layer ResNet, #Paras=0.46M2021.03 | 64.33 | |
| BiCBackbone=modified 32-layer ResNet, #Paras=0.46M2021.03 | 63.8 | |
| iCaRL2022.04 | 63.5 | |
| COIL2022.04 | 62.98 | |
| BiC2022.04 | 62.37 | |
| iCaRLBackbone=modified 32-layer ResNet, #Paras=0.46M2021.03 | 61.16 | |
| PODNet2022.04 | 47.87 |