Class-Incremental Learning on ImageNet-1K (test)
69.46Avg AccuracyBYOL (Joint)
Evaluation Results
| Method | Links | |
|---|---|---|
| BYOL (Joint)Framework=BYOL, Strategy=Joint, Number of Tasks=5T2023.06 | 69.46 | |
| BYOL + PNRFramework=BYOL, Strategy=Pseudo-Negative Regularization, Number of Tasks=5T2023.06 | 66.12 | |
| BYOL + CaSSLeFramework=BYOL, Strategy=CaSSLe, Number of Tasks=5T2023.06 | 64.78 | |
| BYOL + PNRFramework=BYOL, Strategy=Pseudo-Negative Regularization, Number of Tasks=10T2023.06 | 62.56 | |
| BYOL + CaSSLeFramework=BYOL, Strategy=CaSSLe, Number of Tasks=10T2023.06 | 61.93 | |
| BYOL (FT)Framework=BYOL, Strategy=Fine-Tuning, Number of Tasks=5T2023.06 | 60.76 | |
| MoCo (Joint)Framework=MoCo, Strategy=Joint, Number of Tasks=5T2023.06 | 60.62 | |
| BYOL (FT)Framework=BYOL, Strategy=Fine-Tuning, Number of Tasks=10T2023.06 | 57.15 | |
| MoCo + PNRFramework=MoCo, Strategy=Pseudo-Negative Regularization, Number of Tasks=5T2023.06 | 56.87 | |
| MoCo + PNRFramework=MoCo, Strategy=Pseudo-Negative Regularization, Number of Tasks=10T2023.06 | 55.88 | |
| NAPA-VQNumber of incremental tasks (T)=52023.08 | 55.11 | |
| NAPA-VQNumber of incremental tasks (T)=102023.08 | 53.04 | |
| MoCo + CaSSLeFramework=MoCo, Strategy=CaSSLe, Number of Tasks=5T2023.06 | 50.57 | |
| MoCo (FT)Framework=MoCo, Strategy=Fine-Tuning, Number of Tasks=5T2023.06 | 48.33 | |
| NAPA-VQNumber of incremental tasks (T)=202023.08 | 45.46 | |
| MoCo + CaSSLeFramework=MoCo, Strategy=CaSSLe, Number of Tasks=10T2023.06 | 43.38 | |
| MoCo (FT)Framework=MoCo, Strategy=Fine-Tuning, Number of Tasks=10T2023.06 | 42.55 |