BYOL-Pytorch

Pytorch Implementation of BYOL: Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning (https://arxiv.org/abs/2006.07733).
Major part of Code is inspired from https://github.com/sthalles/PyTorch-BYOL.
The Code has more appropriate Naming Convention.

Contents

Implementation has both Notebook version and also a .py version and also Google Colab.

Google Colab Version

Ready to run Colab Version of BYOL is available at BYOL-Pytorch.

Default Training

Running the Python File without any changes trains BYOL with CIFAR10 Dataset.
All the Parameters are contained in Params Object in the script.

Custom Training

Change the Dataset Object.
Update the Required Parameters in the Params Object.

Feedback

Any feedback can be raised as an Issue and Pull Requests can be made.