/libself

PyTorch Lightning based framework to run experiments for self-supervised learning tasks.

Primary LanguagePythonMIT LicenseMIT

LIBSELF - LIghtning Based SElf-supervised Learning Framework

Introduction

This is a PyTorch Lightning based framework to run experiments for self-supervised learning tasks. We borrow structure and some code from SSLIME framework (mentioned in references) but use PyTorch Lightning's experiment pipeline to reduce boilerplate code and enable easier logging and multi-GPU, multi-node and mixed precision training.

Getting Started

coming soon...

License

This framework is released under the MIT license, as found in the LICENSE file.

References

  • Bhooshan, Suvrat and Misra, Ishan and Fergus, Rob and Goyal, Priya. "SSLIME: A Toolkit for Multi-Modal Self-Supervised Training and Benchmarking" (2019)
  • Falcon, W.A. et al. "PyTorch Lightning" (2019)