ImageNet-Cartoon & ImageNet-Drawing

Benchmarking the robustness to distribution shifts traditionally relies on dataset collection which is typically laborious and expensive, in particular for datasets with a large number of classes like ImageNet. An exception to this procedure is ImageNet-C, a dataset created by applying common real-world corruptions at different levels of intensity to the (clean) ImageNet images. Inspired by this work, we introduce ImageNet-Cartoon and ImageNetDrawing, two datasets constructed by converting ImageNet images into cartoons and colored pencil drawings, using a GAN framework [1] and simple image processing [2], respectively. We show that the accuracy of pretrained ImageNet models decreases significantly on the proposed datasets.

ImageNet-Cartoon

Images are taken from the ImageNet dataset and the transformed into cartoon using the GAN framework proposed by [1].

ImageNet-Drawing

Images are taken from the ImageNet dataset and the transformed into colored pencil drawings using the simple image processing described in [2].

Access

We uploaded the dataset on Zenodo, as well as the source code on Github to generate the two datasets.

License

We released the data with the Creative Commons Attribution 4.0 International.

References

[1] Wang, X. and Yu, J. Learning to cartoonize using white-box cartoon representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020.

[2] Lu, C., Xu, L., and Jia, J. Combining Sketch and Tone for Pencil Drawing Production. In Asente, P. and Grimm, C. (eds.), International Symposium on Non-Photorealistic Animation and Rendering. The Eurographics Association, 2012. ISBN 978-3-905673-90-6. doi: 10.2312/PE/NPAR/NPAR12/065-073.

Citation

You can see the full paper here. Please cite as

@inproceedings{salvador2022imagenetcartoon,
    title={ImageNet-Cartoon and ImageNet-Drawing: two domain shift datasets for ImageNet},
    author={Tiago Salvador and Adam M Oberman},
    booktitle={ICML 2022 Shift Happens Workshop},
    year={2022},
    url={https://openreview.net/forum?id=YlAUXhjwaQt}
}