Self-Supervised Learning of Image Features with SwAV (with author Mathilde Caron)

In this video, author Mathilde Caron features “SWaV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments”.

SwAV is the latest paper from FAIR and Inria to post state of the art results in self-supervised learning. The paper combines ideas from contrastive learning and clustering based approaches to train image representations. The swapped prediction problem loss, used in SwAV, enforces consistency in cluster assignment of representations between different views of the same image.

Check out our other video featuring deep dive to the SwAV loss function: https://youtu.be/M_DgS3XGeJc
And step-by-step PyTorch Lightning implementation of SwAV: https://youtu.be/5irer8A2HoY

You can find our implementation in Lightning Bolts, a deep learning research toolkit with SOTA models: https://pytorch-lightning-bolts.readthedocs.io/en/latest/self_supervised_models.html#swav

Paper: https://arxiv.org/abs/2006.09882

GitHub: https://github.com/PyTorchLightning/pytorch-lightning
Lightning Website: https://www.pytorchlightning.ai/
Follow us on Twitter: https://twitter.com/PyTorchLightnin

Source of this PyTorch Lightning Video

AI video(s) you might be interested in …