Class Incremental Learning on CIFAR-100 (Inc5 Split)
45.9Average AccuracyOurs-LWM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Ours-LWMBackbone=Pretrained ResNet-18, Buffer=02026.06 | 45.9 | 31.65 | |
| Ours-LWM+ConBackbone=Pretrained ResNet-18, Buffer=02026.06 | 45.86 | 31.64 | |
| DERBackbone=ResNet32, Buffer=20002026.06 | 40.43 | 32.75 | |
| iCaRLBackbone=ResNet32, Buffer=20002026.06 | 40.05 | 28.72 | |
| BiCBackbone=ResNet32, Buffer=20002026.06 | 28.02 | 22.48 | |
| PODNetBackbone=Cosine ResNet32, Buffer=20002026.06 | 19.69 | 9.98 | |
| LwFBackbone=ResNet32, Buffer=02026.06 | 19.33 | 9.76 | |
| Fine-tuningBackbone=Pretrained ResNet-18, Buffer=02026.06 | 16.37 | 4.55 | |
| SimpleCILBackbone=Cosine ResNet32, Buffer=02026.06 | 11.67 | 4.52 | |
| FeTrILBackbone=ResNet32, Buffer=02026.06 | 10.56 | 4.42 |