This project hosts the code for implementing the ALICE algorithm for few-shot class-incremental classification, as presented in our paper:
Can Peng, Kun Zhao, Tianren Wang, Meng Li, Brian C. Lovell; In: ECCV 2022.
main_base.py: training for the base task.
main_inc_ncm.py: evaluation for the base and incremental tasks.
run_base_CIFAR100.sh: config and dataset settings for base session model trained on CIFAR100 dataset.
run_base_CUB200.sh: config and dataset settings for base session model trained on CUB200 dataset.
run_base_miniImageNet.sh: config and dataset settings for base session model trained on miniImageNet dataset.
run_inc_ncm_CIFAR100.sh: config and dataset settings for incremental session evaluation on CIFAR100 dataset.
run_inc_ncm_CUB200.sh: config and dataset settings for incremental session evaluation on CUB200 dataset.
run_inc_ncm_miniImageNet.sh: config and dataset settings for incremental session evaluation on miniImageNet dataset.
To perform experiments on different incremental sessions, please modify the input arguments - "num_cls" and "current_session" accordingly.
To perform experiments on different few-shot settings (5-shot or 1-shot), please modify the input arguments - "used_img" and "balanced" accordingly.
Our project references the codes in the following repos. We thank the authors for making their code public.