🎉 Big news: LightlyTrain now supports DINOv2. Read our announcement.

Smart Data Selection for Computer Vision

Find the data that matters, save costs and improve model performance

LightlyOne includes the following components

Data Selection Engine

Automatically find the most valuable data for labelling.
Use algorithmic methods to identify and select this data.
Leverage active learning to improve model training efficiency.

Data Viewer & Report

Explore your dataset and create subsets for training.
Show statistics for each selection job submitted to the engine.
Make sure your selection strategy and configuration work well.

Lightly API

Automate your whole data selection pipeline.
Provides authentication, access management, and collaboration functionalities.
Enables you to share data within your team.

Build State-of-Art ML Pipelines With LightlyOne

Lightly selects the subset of your data with the biggest impact on model accuracy, allowing you to improve your model iteratively by using the best data for retraining.

Why Lightly?

14.6x

Increased mAP

90%

Decreased labeling costs

Productionized

Reliable, scalable & optimized for production

No Guesswork

Consistent, traceable & reproducible approach

Smart

State-of-the-art algorithms & embeddings

Low effort

Easy use, integrates tech stack, fully automated

Improve Model Performance While Reducing Labeling Costs

Select the right data with our configuration suite

Embeddings

Selection based on how similar/diverse images are

Diversity

Select diverse objects or images

Similarity search

Find similar objects or images

Metadata

Selection based on collected metadata

Metadata thresholding

Metadata thresholding

Metadata balancing

Balance images across cities

Predictions

Selection based on predictions and their probabilities

Active learning

Find weaknesses of your model

Object balancing

Oversample rare objects

Easy integration into your ML pipeline

LightlyOne Python SDK

Easily schedule a new run with powerful selection strategies with our LightlyOne Python Client

LightlyOne Worker

The run is picked up by the LightlyOne Worker within your own infrastructure and securely interacts with your data in your preferred cloud or local storage

Pipeline integration

The LightlyOne API allows you to easily integrate with external tools.

Seamless integration with other MLOps tools

Data Selection

Labeling

Label QA

Model tooling

Model QA

Data Storage

Local Storage

Experience LightlyOne to optimize your data pipeline.

Improve your model accuracy by using the best data for retraining.