/RotNet

Unsupervised representation learning by predicting image rotation

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

Deep Learning Project (M2 Data Science 2022/2023)

Digit recognition of MNIST dataset (but few images are labeled, ie. 10 per number/ chosen randomly) thanks to a self-supervised approach by predicting image rotation.

  • Pretrained models are available in the /save/models folder.
  • Preprocessing, models training and evaluations are explained in the /notebooks folder.
  • A French version of the report is accessible in the /docs folder.

References

Spyros Gidaris, Praveer Singh and Nikos Komodakis (2018) Unsupervised Representation Learning by Predicting Image Rotations, ICLR, arXiv:1803.07728

License

Copyright © 2023 Data Saiyentist.
This project is GNU General Public License v3.0 licensed.