Case Study: Captur achieves scalable real-time edge computer vision with Weights & Biases

Captur deploys mobile computer vision models under 20MB with Weights & Biases

Captur, a company providing edge AI for real-time image verification in last-mile delivery and micromobility, faced significant challenges developing computer vision models that could run instantly and accurately on diverse mobile devices with strict size and power constraints. To manage their complex machine learning workflow and overcome these obstacles, the team turned to the Weights & Biases platform.

By integrating Weights & Biases, particularly the Artifacts and Registry features, Captur established a streamlined model lifecycle. This enabled automated evaluations, one-click deployments, and a clear release candidate process, leading to a more efficient and reliable deployment cycle. The solution provided a central system of record for all ML activities, allowing the team to confidently achieve specific model improvement goals, such as a targeted 15% PR-AUC increase for hazard detection.

Captur