Awesome

EFT paper

Efficient Feature Transformations for Discriminative and Generative Continual Learning: CVPR-21

Class Incremental Learning

CIFAR-100/10 and CIFAR100/20

Table-1 Result

Jupyter file cifar100-10_816_class_incremental.ipynb train and evaluate the result for the cfiar100 dataset, divided into 10 task and 10 classes each.

Here: groupwise conv (gp)=8 and pointwise conv (pt)=16

change the value of gp and pt for the other results from table-1

Figure-3 result

Jupyter file cifar100-20_816_class_incremental.ipynb train and evaluate the result for the cfiar100 dataset, divided into 20 task and 5 classes each.

Jupyter file cifar100-5_816_class_incremental.ipynb train and evaluate the result for the cfiar100 dataset, divided into 5 task and 20 classes each.

Task Incremental Learning

Tiny ImageNet 200 on VGG-16 architecture

Table-2 Result

Jupyter file tiny_imagenet_200-10_vgg16-3conv.ipynb divide the 200 classes into 10 task, 20 class each and evaluate the result.