/DDE_CIL

Primary LanguagePython

CVPR2021 Incremental Learning

[Paper]

This repository is for the paper "Distilling Causal Effect of Data in Class-Incremental Learning".


Instructions

  1. Dependencies

    • Python 3.6 (Anaconda3 Recommended)
    • Pytorch 0.4.0
    • torchvision 0.2.1
    • numpy 1.18.1
  2. Getting Started

    • the data for CIFAR100 and ImageNet are put in cifar100-class-incremental/data and imagenet-class-incremental/data, or you can make soft links to the directories which include the corresponding data
    • make soft links for utils_incremental folder under cifar100-class-incremental and imagenet-class-incremental
    • make folders logs, results and checkpoint under cifar100-class-incremental and imagenet-class-incremental
    • see cifar100-class-incremental/run.sh for the experiments on CIFAR100
    • see imagenet-class-incremental/run.sh for the experiments on ImageNet-Subset
    • see imagenet-class-incremental/run_all.sh for the experiments on ImageNet-Full

Citation

Please cite the following paper if you find this useful in your research:

@InProceedings{Hu_20121_CVPR,
author = {Hu, Xinting and Tang, Kaihua and Miao, Chunyan and Hua, Xian-Sheng and Zhang, Hanwang},
title = {Distilling Causal Effect of Data in Class-Incremental Learning},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021}
}