Lightly provides a data curation platform that utilizes self-supervised learning and active learning techniques to optimize the selection of training data for machine learning models. By reducing data redundancy and bias, Lightly enables companies to achieve up to 92% lower labeling costs and improve model accuracy by 19%.
Funding
$4.2M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Training robust machine learning models, especially in computer vision, requires large, diverse, and well-labeled datasets, which can be expensive and time-consuming to acquire and curate. Data redundancy and bias within these datasets can negatively impact model performance and generalization capabilities.
Solution
Lightly offers a data curation platform that leverages self-supervised learning and active learning techniques to optimize training datasets for machine learning models. The platform identifies and selects the most valuable data points, reducing redundancy and bias while improving model accuracy. Lightly provides tools for smart data capturing on device (Lightly Edge), data selection and viewing (Lightly One), and self-supervised pretraining (Lightly Train), enabling users to build efficient and effective data pipelines. By integrating with existing ML stacks, Lightly automates data curation runs and facilitates closed-loop feedback cycles for continuous model improvement.
Target Audience
Lightly's primary customers are enterprises, researchers, and startups in the computer vision, autonomous driving, smart farming, and generative AI fields who need to optimize their data pipelines for machine learning.
Features
- **Lightly Edge:** Smart data capturing on-device to find the most valuable data directly at the source.
- **Lightly One:** Data selection and viewing tools to gain insights and implement optimal selection strategies.
- **Lightly Train:** Self-supervised learning capabilities to pretrain models and improve performance.
- Data insights to understand data distribution, mitigate bias, and identify edge cases.
- Smart data selection using model predictions, embeddings, and metadata.
- Automated pipeline management for data curation runs and tracking new data.
- Seamless integration with various data storage, labeling, and model tooling providers.