Documentation
See also
These pages document the Lightly self-supervised learning research library. If you are instead looking to leverage SSL and distillation pretrain with only a few lines of code head over to LightlyTrain instead. And if you are looking for the LightlyOne Worker Solution with advanced active learning algorithms and selection strategies to select the best samples within millions of unlabeled images or video frames stored in your cloud storage or locally, please follow our LightlyOne Worker documentation.
LightlySSL is a computer vision framework for self-supervised learning and latent world models.
World Models
A world model learns how a scene responds to actions. LeWM learns it from raw frames and actions, then plans by imagining the outcome of each move in its own embedding space.
Each agent (orange) gets only a picture of its goal (white box). LeWM finds a way through the door. It learned from random play: no rewards, no labels, no map of the room.
Start with Train a World Model and Plan with It.
With LightlySSL you can train deep learning models using self-supervision. This means, that you don’t require any labels to train a model. LightlySSL has been built to help you understand and work with large unlabeled datasets. It is built on top of PyTorch and therefore fully compatible with other frameworks such as Fast.ai.
For a commercial version with more features, including Docker support and pretraining models for embedding, classification, detection, and segmentation tasks with a single command, please contact sales@lightly.ai.
Lightly AI
Discord (We have weekly paper sessions!)
Getting Started
World Models
Tutorials
- Tutorial 1: Structure Your Input
- Tutorial 2: Train MoCo on CIFAR-10
- Tutorial 3: Train SimCLR on Clothing
- Tutorial 4: Train SimSiam on Satellite Images
- Tutorial 5: Custom Augmentations
- Tutorial 6: Pre-train a Detectron2 Backbone with Lightly
- Tutorial 7: Finetuning Lightly Checkpoints
- Tutorial 8: Using timm Models as Backbones
Examples
Python API
On-Premise