Class-Incremental Learning on ImageNet 500 base + 10x50 novel classes
70.67Accuracy (t=0..10)score fusion (best-Acc_avg)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| score fusion (best-Acc_avg)Backbone=ResNet182022.04 | 70.67 | 61.06 | |
| score fusion (best-balanced)Backbone=ResNet182022.04 | 69.4 | 65.95 | |
| score fusion (best-Acc_avg)Backbone=ResNet102022.04 | 66.23 | 55.31 | |
| score fusion (best-Acc_all)Backbone=ResNet182022.04 | 65.66 | 67.48 | |
| score fusion (best-balanced)Backbone=ResNet102022.04 | 65.34 | 61.03 | |
| PODNetBackbone=ResNet182022.04 | 64.13 | — | |
| score fusion (best-Acc_all)Backbone=ResNet102022.04 | 61.45 | 62.7 | |
| BiCBackbone=ResNet182022.04 | 44.31 | — |