class-incremental-learning/mnemonics-training

Mnemonics Training

[PDF](https://arxiv.org/pdf/2002.10211.pdf)

Requirements

See the versions for the requirements here.

Download the Datasest

See the details here.

Running Experiments

Running experiments for baselines

cd ./mnemonics-training/1_train
python main.py --method=baseline --nb_cl=10
python main.py --method=baseline --nb_cl=5
python main.py --method=baseline --nb_cl=2

Running experiments for our method

cd ./mnemonics-training/1_train
python main.py --method=mnemonics --nb_cl=10
python main.py --method=mnemonics --nb_cl=5
python main.py --method=mnemonics --nb_cl=2

Performance

Average accuracy (%)

Method	Dataset	5-phase	10-phase	25-phase
LwF	CIFAR-100	52.44	48.47	45.75
LwF w/ ours	CIFAR-100	54.21	52.72	51.59
iCaRL	CIFAR-100	58.03	53.01	48.47
iCaRL w/ ours	CIFAR-100	60.01	57.37	54.13

Forgetting rate (%, lower is better)

Method	Dataset	5-phase	10-phase	25-phase
LwF	CIFAR-100	45.02	42.50	39.86
LwF w/ ours	CIFAR-100	40.00	36.50	34.25
iCaRL	CIFAR-100	32.87	32.98	36.32
iCaRL w/ ours	CIFAR-100	25.93	26.92	28.92

Citation

Please cite our paper if it is helpful to your work:

@inproceedings{liu2020mnemonics,
author    = {Liu, Yaoyao and Su, Yuting and Liu, An{-}An and Schiele, Bernt and Sun, Qianru},
title     = {Mnemonics Training: Multi-Class Incremental Learning without Forgetting},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
pages     = {12245--12254},
year      = {2020}
}

Acknowledgements

Our implementation uses the source code from the following repositories:

• Learning a Unified Classifier Incrementally via Rebalancing

• iCaRL: Incremental Classifier and Representation Learning

• Dataset Distillation

• Generative Teaching Networks