Title: Semantic Drift Compensation for Class-Incremental Learning.
The paper will be published at the conference of 2020 Computer Vision and Pattern Recognition (CVPR20). An pre-print version is available. Poster is linked.
Lu Yu, Bartłomiej Twardowski, Xialei Liu, Luis Herranz, Kai Wang, Yongmei Cheng, Shangling Jui, Joost van de Weijer
All training and test are done in Pytorch framework.
Pytorch vesion: 0.3.0.post4
Python version: 2.7
Code with higher pytorch version will come soon...
We evaluate our system in several datasets, including CUB-200-2011, Flowers-102, Caltech-101, CIFAR100, ImageNet-Subset(the first 100 classes of full ImageNet).
Please download CUB-200-2011 , Flowers-102, Caltech-101, CIFAR100 and ImagNet-Subset.(Note: some datasets do not split the train set and test set in the original folder, the splited datasets can be download from this link according to the original provided train/test text file.)
The loss functions in the code refer to source repository.