How Fox Robotics Reduced Annotation
Time From Months to Hours
Labeling and reviewing images by hand proved to be a challenging assignment for the small team. The company found that the sheer number of annotated field and crop data they needed to train their ML models were not scalable in-house.
Superb AI’s personalized data services in conjunction with custom automated labeling and active learning workflows.
>72% reduction in annotation costs
>20x faster labeling compared to manual annotation
Significant improvement in label accuracy
Fox Robotics is a pioneering automation solution provider that focuses primarily on developing agricultural robots for soft fruit farms. The company aims to automate routine and hard jobs, boosting farm productivity and crop yields despite a shrinking workforce. To allow farmers to spend more time on farming, the company sought new data labeling methods to avoid compromising quality, getting stuck in pre-production, or overspending on external labelers that lacked sufficient subject matter expertise.
The team's initial approach to data labeling, using in-house resources and third-party labeling tools to annotate complex images, proved too slow, and the quality of outputs was inconsistent. Automated data labeling seemed promising, but most options they explored couldn’t handle nuanced data and lacked built-in QA mechanisms. The team at Fox Robotics quickly realized that they needed a partner to help them label data faster, with increased accuracy, precision, and recall, and to adopt technology that could scale with future growth.
“As a practitioner, I think the feature that really stood out to me is their custom auto-label. The amount of quality labels you can get in such a short time is nothing short of wild. The suite was also easy to learn and use, from exporting data to manual QA.”
That’s when Fox Robotics discovered Superb AI and the concept of data preparation automation in the form of Custom Auto-Label, which provides custom automation tailored to a company’s specific use case. The team also turned to Superb AI’s personalized data services, which annotated most ground truth labels used to train the automated data labeling models, managed the full project end-to-end, and provided guidance and advice during key stages of each project. By establishing this partnership between services and technology, Fox Robotics was able to eliminate cumbersome manual labeling in-house while also avoiding expensive long-term contracts with third-party labeling services.
In terms of results, adopting Superb AI’s technology, guided in large part by the data services team, has reduced Fox Robotics' labeling costs by as much as 72% while also significantly improving label accuracy and consistency. Without compromising on speed or accuracy, the team has accelerated their time to market and developed more features for their AgTech robotic solutions. In the future, the team plans to tackle new and more-complex use cases while using tools like custom auto-label to scale labeling whenever needed.