iirc 1.0.0

This package provides a way for adapting the different datasets (currently supports *CIFAR-100* and *ImageNet*) to the *iirc* setup and the *class incremental learning* setup, and loading them in a standardized manner.

iirc package

This package provides a way for adapting the different datasets (currently supports CIFAR-100 and ImageNet) to the iirc setup and the class incremental learning setup, and loading them in a standardized manner.

The documentation and usage guide are available here

Homepage | Paper | Documentation

Installation

you can install this package using the following command

pip install iirc

Dataset Downloading Instructions

CIFAR-100

To be able to run the code with CIFAR-100 derived datasets, just download the dataset from the official website and extract it, or use the ./utils/download_cifar.py file.

ImageNet

In the case of ImageNet, it has to be downloaded manually, and be arranged in the following manner:

• dataset folder
  • train
    • n01440764
    • n01443537
    • …
  • val
    • n01440764
    • n01443537
    • …

Contributing

If you think you can help us make the iirc package more useful for the lifelong learning community, please don't hesistate to submit an issue or send a pull request.

Citation

If you find this work useful for your research, this is the way to cite it:

@article{abdelsalam2020iirc,
  Author = {Mohamed Abdelsalam and Mojtaba Faramarzi and Shagun Sodhani and Sarath Chandar},
  Title = {IIRC: Incremental Implicitly-Refined Classification},
  Year = {2020},
  arxiv = {https://arxiv.org/abs/2012.12477}
}