Learning without Forgetting Implementation

I am currently working on replication of LwF CL method, does any of you have tried replicating this method? Using 2 different datasets as a set of tasks. For example Imagenet for first task and CUB dataset for second tasks? I would really love to have help Thank you.