In order to facilitate reproducibility of the experiments in Class-incremental learning: survey and performance evaluation (arxiv), scripts for the main experiments are provided. Their main usage is as follows:
./script_cifar100.sh <approach> <gpu> <scenario> [<results_dir>]where:
<approach>- approach to be used, from the ones in./src/approaches/<gpu>- index of GPU to run the experiment on<scenario>- specific rehearsal scenario (base: no exemplars, fixd: fixed memory, grow: growing memory)[<results_dir>]- results directory (optional), by default it will be./results