DER: Dynamically Expandable Representation for Class Incremental Learning

Benchmark Model Rank Results
incremental-learning-on-cifar-100-50-classes-2DER(Standard ResNet-18)#3Average Incremental Accuracy: 72.45
incremental-learning-on-cifar-100-50-classes-2DER(Modified ResNet-32)#7Average Incremental Accuracy: 66.36
incremental-learning-on-cifar-100-50-classes-3DER(Standard ResNet-18)#3Average Incremental Accuracy: 72.60
incremental-learning-on-cifar-100-50-classes-3DER(Modified Res-32)#7Average Incremental Accuracy: 67.60
incremental-learning-on-cifar-100-50-classes-4DER (w/o P)#3Average Incremental Accuracy: 74.61
incremental-learning-on-cifar-100-b0-5stepsDER(w/o P)#4Average Incremental Accuracy: 76.80
incremental-learning-on-cifar100-b0-10stepsDER(ResNet-18)#4Average Incremental Accuracy: 74.64
incremental-learning-on-cifar100b020stepDER(ResNet-18)#4Average Incremental Accuracy: 73.98
incremental-learning-on-imagenet-10-stepsDER w/o Pruning#2Average Incremental Accuracy: 68.84Final Accuracy: 60.16
incremental-learning-on-imagenet-10-stepsDER#5Average Incremental Accuracy: 66.73Final Accuracy: 58.62
incremental-learning-on-imagenet-100-50-2DER#2Average Incremental Accuracy: 77.73
incremental-learning-on-imagenet100-10-stepsDER w/o Pruning#5Average Incremental Accuracy: 77.18Final Accuracy: 66.70
incremental-learning-on-imagenet100-10-stepsDER#7Average Incremental Accuracy: 76.12Final Accuracy: 66.07