Lightly AG

Lightly AG builds data-centric AI infrastructure for computer vision. The company develops tools that help machine learning teams curate, label and manage visual datasets at scale, with the aim of reducing the cost and time required to train production-ready models. Its work spans computer vision, self-supervised learning, representation learning, data curation and dataset management. The systems are used in industries including autonomous driving, medical imaging and industrial inspection, and the company serves customers globally.

Lightly's product line addresses different stages of the data pipeline. LightlyOne identifies the most valuable data samples for labeling, reducing annotation cost and time. LightlyTrain provides self-supervised pretraining frameworks intended to remove the need for expensive manual annotation. Two further projects are released as open source: LightlySSL, a library for self-supervised learning, and LabelFormat, a standardised data annotation format. Taken together, the products are designed to enable more accurate computer vision systems with significantly less labelled data.

The company was spun out of ETH Zurich and is backed by Y Combinator. Its stated approach combines research with practical deployment tools, and a portion of its software is developed openly on an open-source basis.