Task Incremental Learning on ImageNet-1K
95Top-5 Acc (T1)EEIL
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| EEILReplay-based approach=true2021.03 | 95 | 95.5 | 86 | 77.5 | 71 | 68 | 62 | 59.8 | 55 | 52 | |
| CCLLExtra parameters per task=around 0.5%, Extra FLOPS=0.98%2021.03 | 91.4 | 88.3 | 86.5 | 86.6 | 84.6 | 83.5 | 82.7 | 81.7 | 81.2 | 81.3 | |
| RPS-NetReplay-based approach=true2021.03 | 90.2 | 88.4 | 82.4 | 75.9 | 66.9 | 62.5 | 57.2 | 54.2 | 51.9 | 48.8 | |
| RKRExtra parameters per task=around 0.5%, Extra FLOPS=2.8 × 1e-4%2021.03 | 90.2 | 88.7 | 88.1 | 88.2 | 86.6 | 85.7 | 85 | 84.2 | 83.8 | 84.4 | |
| iCaRLReplay-based approach=true2021.03 | 90.1 | 82.8 | 76.1 | 69.8 | 63.3 | 57.2 | 53.5 | 49.8 | 46.7 | 44.1 |