arthurdouillard/incremental_learning.pytorch
arthurdouillard/incremental_learning.pytorch has 411 GitHub stars and 57 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-21 14:38 UTC. Last push 2023-10-03.
計測したスター増加
- 24時間
- 0
- 7日間
- 履歴を収集中
- 15日間
- 履歴を収集中
- 1か月
- 履歴を収集中
期間ごとの純増数です。期間は重複しており、日別の値ではありません。未計測の値は空欄です。スター数は注目度の指標であり、品質の保証ではありません。
概要
A collection of incremental learning paper implementations including PODNet (ECCV20) and Ghost (CVPR-W21).
機械学習README.md
オリジナルIncremental Learners for Continual Learning
Repository storing some my public works done during my PhD thesis (2019-).
You will find in there both known implementation (iCaRL, etc.) but also all my papers. You can find the list of the latter on my Google Scholar.
My work on continual segmentation can be found here and on continual data loaders here.
Structures
Every model must inherit inclearn.models.base.IncrementalLearner.


If you use this paper/code in your research, please consider citing us:
@inproceedings{douillard2020podnet,
title={PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning},
author={Douillard, Arthur and Cord, Matthieu and Ollion, Charles and Robert, Thomas and Valle, Eduardo},
booktitle={Proceedings of the IEEE European Conference on Computer Vision (ECCV)},
year={2020}
}
To run experiments on CIFAR100 with three different class orders, with the challenging setting of 50 steps:
python3 -minclearn --options options/podnet/podnet_cnn_cifar100.yaml options/data/cifar100_3orders.yaml \
--initial-increment 50 --increment 1 --fixed-memory \
--device <GPU_ID> --label podnet_cnn_cifar100_50steps \
--data-path <PATH/TO/DATA>
Likewise, for ImageNet100:
python3 -minclearn --options options/podnet/podnet_cnn_imagenet100.yaml options/data/imagenet100_1order.yaml \
--initial-increment 50 --increment 1 --fixed-memory \
--device <GPU_ID> --label podnet_cnn_imagenet100_50steps \
--data-path <PATH/TO/DATA>
And ImageNet1000:
Likewise, for ImageNet100:
python3 -minclearn --options options/podnet/podnet_cnn_imagenet100.yaml options/data/imagenet1000_1order.yaml \
--initial-increment 500 --increment 50 --fixed-memory --memory-size 20000 \
--device <GPU_ID> --label podnet_cnn_imagenet1000_10steps \
--data-path <PATH/TO/DATA>
Furthermore several options files are available to reproduce the ablations showcased
in the paper. Please see the directory ./options/podnet/ablations/.

If you use this paper/code in your research, please consider citing us:
@inproceedings{douillard2020ghost,
title={Insight From the Future for Continual Learning},
author={Arthur Douillard and Eduardo Valle and Charles Ollion and Thomas Robert and Matthieu Cord},
booktitle={arXiv preprint library},
year={2020}
}
The code is still very dirty, I'll clean it later. Forgive me.
関連プロジェクトを探す
読み込み中…