Incremental Learning 벤치마크
Incremental Learning on ImageNet-100 - 50 classes + 5 steps of 10 classes
Average Incremental Accuracy
- 2016-11-23 — iCaRL*: Average Incremental Accuracy 65.56
- 2020-04-28 — PODNet: Average Incremental Accuracy 75.82
- 2021-02-18 — CCIL-SD: Average Incremental Accuracy 79.44
- 2022-04-10 — FOSTER: Average Incremental Accuracy 80.22
| Rank | Model | Average Incremental Accuracy | Paper | Code | Year |
|---|---|---|---|---|---|
| 1 | FOSTER | 80.22 | FOSTER: Feature Boosting and Compression for Class-Incremental Learning | g-u-n/pycil · G-U-N/ECCV22-FOSTER | 2022 |
| 2 | RMM (ResNet-18) | 79.52 | RMM: Reinforced Memory Management for Class-Incremental Learning | g-u-n/pycil · aimagelab/mammoth · yaoyaoliu/rmm · +1 | 2023 |
| 3 | CCIL-SD | 79.44 | Essentials for Class Incremental Learning | sud0301/essentials_for_CIL | 2021 |
| 4 | PODNet | 75.82 | PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning | g-u-n/pycil · arthurdouillard/incremental_learning.pytorch | 2020 |
| 5 | iCaRL* | 65.56 | iCaRL: Incremental Classifier and Representation Learning | ContinualAI/avalanche · g-u-n/pycil · aimagelab/mammoth · +7 | 2016 |