MixER: Mixup-based Experience Replay for Online Class-Incremental Learning

Official Code for "MixER: Mixup-based Experience Replay for Online Class-Incremental Learning". Our code is based on the implementation of Mammoth.

Running Experiments

Examples:

CIFAR-10

python main.py --dataset seq-cifar10 --model mixer --buffer_size 500 --lr 0.01 --batch_size 10 --minibatch_size 10 --n_epochs 1 --gamma 5.0

CIFAR-100

python main.py --dataset seq-cifar100 --model mixer --buffer_size 1000 --lr 0.01 --batch_size 10 --minibatch_size 10 --n_epochs 1 --gamma 5.0

Tiny-ImageNet

python main.py --dataset seq-tinyimg --model mixer --buffer_size 2000 --lr 0.01 --batch_size 10 --minibatch_size 10 --n_epochs 1 --gamma 5.0

Cite Our Work

If you find this code useful, please consider reference in our paper:

@InProceedings{lim2024mixer,
    author    = {Lim, Won-Seon and Yu Zhou and Kim, Dae-Won and Lee, Jaesung},
    title     = {MixER: Mixup-based Experience Replay for Online Class-Incremental Learning},
    journal   = {IEEE ACCESS},
    volume    = {12},
    pages    = {41801--41814},
    year      = {2024},
    publisher = {IEEE}
}