Accelerate. Iterate. Celebrate.

Lodestar is the world’s first active learning video annotation platform to accelerate high-quality dataset and model creation.

Accelerate Your Projects

More Labels, Better Datasets, Less Time.

Today, training a computer vision model involves extensive data preparation and curation, and a lot of coding to ingest the data and finally begin labeling. Then data scientists wait weeks for the dataset to be labeled before they can begin validating their model. It can take months of multiple cycles of labeling, exporting, training a model, analyzing the results, and starting again to achieve success.

Lodestar changes all that. An integrated platform for labeling and training with native video support trains a model in real time and allows your teams to work in parallel. Active learning and AI prediction help generate labels and automatically curate your data. You’ll finish in weeks what used to take months.

Why We’re Different

No-Code Labeling

Domain Experts Drive High-Quality Dataset Design

A computer vision model is only as good as the training data used to create it. Rapid iteration guided by data scientists is the key to achieving the best model for your application. Using Lodestar, in-house experts can label industry-specific data without writing any code while data scientists experiment and adjust to increase model accuracy and utility. Solve problems no outsourced workforce, or even in-house data scientists, can fully understand.

Label Faster

Predictive Labeling Increases Productivity

Competitors use general models or a supplied model to generate predictions. Low-accuracy predictions on densely populated frames can make things worse. Only Lodestar continuously improves its prediction model by training it on the actual dataset as it is being labeled. Predictions kick in after just 20 objects have been labeled and keep getting better as the work progresses.

Continuous Training

Active Learning Finds Objects Faster

Active learning uses AI to find objects appearing infrequently in large, sparse datasets, saving time. Other tools use a static, generic model to search the data. Lodestar is built on GPU compute so it can label and train at the same time. It trains a model on the live dataset as labels are added. It uses this custom model to find objects buried in the data that will further accelerate training. We call this “continuous training” and it is only possible by running the whole stack on GPUs.

Optimized for Video

Work with Hours of Video.
No Data Prep, No Upcharges

Add hours of video to a project as your model takes shape. No extra charges or add-on modules and no converting frames to images and searching for the best ones to label. Lodestar is designed for video down to its core. GPU power allows it to process hours of video to drive predictive labeling and active learning.

Lodestar is a Game Changer

“Typically we’d spend up to two months with a client on a proof of concept; labeling video data sets and developing a computer vision model to the point they were confident it would meet their needs and the project would be successful. Using Lodestar Navigator we’re getting there in a matter of hours. It’s not only accelerating our projects and increasing the return on investment, it’s changing the way we approach clients on these projects—we can spend an hour with them and demonstrate results instead of talking about possibilities.”

— Product Manager, IoT consulting firm