Case Study

SnapCalorie Builds Consumer Mobile App with Computer Vision at the Core

Choosing the Best Tools to Build a Computer Vision Business

SnapCalorie, a YCombinator-backed startup, enables users to take a picture of any meal and their algorithm will show the calories, fat, carbs, and protein. This simple end user experience is made possible only with computer vision. The team behind the company co-founded Google Lens and Google Cloud Vision API and they co-authored a peer reviewed paper demonstrating the first algorithm to outperform humans when visually estimating calories and nutrition in a meal. This team of computer vision experts realized they needed best-in-class tools to build their business and decided buying, rather than building, was the best path forward.

Fast and Scalable Software for Annotation and Labeling

SnapCalorie selected Roboflow for dataset management, labeling, annotation, and managing label-only users from outside their organization. The depth of Roboflow Annotate’s functionality is what the computer vision experts needed to manage the pipeline of uploading, searching, assigning, reviewing, and approving annotations. AI-assisted labeling using pre-trained models leverages high-performance compute by AWS to quickly train and update models. Fast labeling workflows like keyboard shortcuts, intelligent defaults, and integrated label creation helped the team to annotate thousands of images quickly and accurately. Secure role-based access keeps the team’s data safe while adding labeling capacity from outside of their organization and project management is made easy with progress views, labeling instructions, and notifications for external labeling users. With Roboflow, the team is able to quickly add new data to train and improve their model as their dataset expands.

Speed increase in model prototyping
Decrease in labeling time per image