This is the official implementation of the paper "LifeLonger: A Benchmark for Continual Disease Classification, MICCAI, 2022".

Please be at the root of the project:

mkdir data; cd data

Create a new directory for each dataset

mkdir bloodmnist (or tissuemnist or pathmnist or organamnist)

Come back the root directory of the project to create a results folder pooling all of the final resutls here.

cd ..
mkdir resutls

Go to scripts folder, run following commands. Each one, run the 5 different baselines on one benchmarks. Here, we train 5 different baselines (ewc, icarl, bic, lucir, finetuning) on bloodmnist dataset. In each command, 0 shows the gpu number. You can adopt following commands for your assigned banchmarks by change the code as: bash ./script_{benchmark_name}.sh ewc 0 fixd ../resutls/benchmark_name where benchmark_name can be values in [bloodmnist, tissuemnist, pathmnist, organamnist)]

cd scripts
bash ./script_bloodmnist.sh ewc 0 fixd ../resutls/bloodmnist
bash ./script_bloodmnist.sh icarl 0 fixd ../resutls/bloodmnist
bash ./script_bloodmnist.sh lwf 0 fixd ../resutls/bloodmnist
bash ./script_bloodmnist.sh eeil 0 fixd ../resutls/bloodmnist
bash ./script_bloodmnist.sh mas 0 fixd ../resutls/bloodmnist
bash ./script_bloodmnist.sh finetuing 0 fixd ../resutls/bloodmnist

for croos domain incremental learning:

bash ./script_cross_domain.sh lwf 0 fixd ../resutls/cross_domain
bash ./script_cross_domain.sh ewc 0 fixd ../resutls/cross_domain
bash ./script_cross_domain.sh icarl 0 fixd ../resutls/cross_domain