Class-Incremental Learning on ImageNet-100 20-task setting (test)
51.08Average Mean AccuracyTagFex + Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TagFex + OursNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 51.08 | 35.78 | |
| DER + OursNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 47.8 | 29.79 | |
| TagFexNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 47.32 | 31.36 | |
| DERNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 43.95 | 25.82 | |
| iCaRL + OursNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 42.54 | 23.9 | |
| MEMO + OursNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 38.76 | 26.16 | |
| FOSTER + OursNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 38.07 | 19.72 | |
| FOSTERNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 35.67 | 15.28 | |
| MEMONumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 34.73 | 22.78 | |
| iCaRLNumber of tasks=20, Replay buffer size=20 exemplars per class2026.03 | 33.45 | 17.62 |