Lightly helped Voxel AI process 50M images in six months, cut retraining time by 50%, double onboarding speed, and improve model accuracy by 10% by selecting only the most relevant unlabeled frames.
Lightly helped Voxel AI process 50M images in six months, cut retraining time by 50%, double onboarding speed, and improve model accuracy by 10% by selecting only the most relevant unlabeled frames.
Voxel AI teaches AI-powered security cameras to detect hazards and inefficiencies in real time. This allows customers to build a proactive safety culture and eliminate near-misses with the ultimate site visibility platform. Voxel’s deep learning-based systems detect various use cases: ergonomics, environment, vehicles, PPE, and operations. For customers,Voxel’s solution does not only offer more security to employees but also financial benefits for the employer by preventing accidents, injuries and alike. Over the past years, the company out of the bay area has matured to one of the world’s leading providers of such systems.
‍
Voxel’s solution must perform on various sites worldwide. Before launch, sample video data is collected from site cameras over a short period of time. But training the AI model based on full recordings of video feeds is neither an efficient nor optimal solution.
For Voxel, it was crucial to find a solution for the following issues:
Voxel was looking for a way to improve the speed of their customer deployments and allow customers to start using their platform quickly by minimizing the amount of time required to train its model. Accordingly, they needed to find an efficient way to select the most interesting data to annotate the most relevant images.
“Lightly enabled us to improve our ML data pipeline in all regards: Selection, Efficiency, and Functionality. This allowed us to cut customer onboarding time by 50% while achieving better model performance.”
Co-Founder/ VP of Engineering
Lightly selects Voxel’s most relevant unlabeled data, utilizing its self-supervised learning framework. This allows Voxel to train its model for each customer more quickly while achieving even higher performance.
Lightly’s active learning feature and multi-factor querying functionality enhanced Voxel’s solutions performance. The integration of Lightly’s labeling tool supported Voxel in streamlining theirMLOps pipeline, increasing efficiency.
In a nutshell, Voxel was able to (1) select the right data to train the model faster and better, (2) learn more about what additional data would enhance model performance, and (3) use this insight to increase their model performance. Thus allowing them to:
No fluff—just results from teams using Lightly to move faster with better data and models.
"Through this collaboration, SDSC and Lightly have combined their expertise to revolutionize the process of frame selection in surgical videos, making it more efficient and accurate than ever before to find the best subset of frames for labeling and model training."
“Lightly gave us transparency to a part of the ML development that is a black box, data. Furthermore, Lightly enabled us to do Active Learning at scale and helped us improve recall and F1-score of our object detector by 32% and 10% compared to our previous data selection method. We finally saw the light in our data using Lightly.”
Data Selection & Data Viewer
Get data insights and find the perfect selection strategy
Self-Supervised Pretraining
Leverage self-supervised learning to pretrain models
Smart Data Capturing on Device
Find only the most valuable data directly on device
See benchmarks comparing real-world pretraining strategies inside. No fluff.
