Member-only story
Lightly is One of the First Open Source Frameworks for Self-Supervised Learning
The library implements some of the most common algorithms and pretrained models for SSL.
I recently started an AI-focused educational newsletter, that already has over 125,000 subscribers. TheSequence is a no-BS (meaning no hype, no news etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. Please give it a try by subscribing below:
Self-Supervised Learning(SSL) is still in a very nascent state, and, as a result, the frameworks and platforms on the market are still quite limited. Lightly is an open-source library for implementing SSL models that have gained strong visibility within the deep learning community. Like other SSL frameworks in these early days, Lightly specializes in computer vision domains. The framework is entirely built on PyTorch and leverages advanced research such as Google’s SimCLR library.
From an architectural standpoint, Lightly provides a modular framework that abstracts the fundamental building blocks of SSL applications. Most Lightly applications are based on the following components:
- Datasets: This component is based on TorchVision datasets and represents the fundamental data unit of Lightly applications.
- Collate Function: This component performs the data augmentation function, which is essential in SSL models.
- Data Loader: This component is based on PyTorch’s data loaders and loads the datasets used to train and test the models.
- Backbone Neural Network: This component represents a pretrained model that can be used in Lightly’s SSL applications.
- Model: This component represents the final SSL program that combines the backbone neural network with other building blocks.
Image credit: Lightly
There are other components of the Lightly framework, but the ones mentioned above encapsulate the key building blocks of any SSL model built using this library. The experience of building SSL models using Lightly can be as simple as a handful of lines of code, as illustrated in the following figure:
The capabilities of the Lightly open-source framework are complemented with the Lightly platform, which provides a complete set of features to manage the lifecycle of datasets used in SSL models. Even though you can get very far with the basic Lightly open-source release, production-grade SSL applications might benefit from the capabilities of the Lightly platform.
