PCR: Proxy-based Contrastive Replay for Online Class-Incremental Continual Learning

Code For CVPR'2023 paper "PCR: Proxy-based Contrastive Replay for Online Class-Incremental Continual Learning"

The framework is based on online-continual-learning.

• CIFAR10 & CIFAR100 will be downloaded during the first run. (datasets/cifar10;/datasets/cifar100)
• Mini-ImageNet: Download from https://www.kaggle.com/whitemoon/miniimagenet/download, and place it in datasets/mini_imagenet/.

PCR is in https://github.com/FelixHuiweiLin/PCR/blob/main/agents/pcr.py.

CIFAR-100

  python general_main.py --num_runs 1 --data  cifar100 --cl_type nc --agent PCR  --retrieve random --update random --mem_size 1000

CIFAR-10

  python general_main.py --num_runs 1 --data cifar10 --cl_type nc --agent PCR --retrieve random --update random --mem_size 200

Mini-Imagenet

python general_main.py --data --num_runs 1  mini_imagenet --cl_type nc --agent PCR --retrieve random --update random --mem_size 1000