Case Study
Rowbot Significantly Boosts Model Performance with Superb AI
Rowbot’s reliance on in-house labeling to produce datasets for corn segmentation and color classification was draining time and resources away from their ML team. They needed to find a way to increase labeling throughput and quality without increasing their workforce.
Superb AI’s personalized data services and labeling tool to receive and review annotated data.
~50% improvement in model performance
>65% decrease in labeling and QA time-per-task
100% delivery on labels on time and to spec
~98% label accuracy rate across 2.4K images
Rowbot specializes in autonomous robotic solutions that can operate between crop rows throughout the growing season regardless of crop height. This enables the precise application of nitrogen fertilizer close to a plant's base for easy uptake. Rowbot is developing new technology (N-Smart™) that will allow their robots to monitor nitrogen stress using a proprietary approach to increase nitrogen use efficiency dramatically. To power N-Smart™ solutions, accurate and extensive datasets are needed, covering various field conditions, lighting, and growth stages. Likewise, the annotations must be very precise and consistent to yield accurate estimates of nitrogen stress, which proved to be very time-consuming.
Before partnering with Superb AI, Rowbot had lots of real-world data from several field tests conducted in Minnesota and Iowa that needed to be annotated. Even though the company had a small in-house team available for data annotation, they couldn’t match the scale or throughput required, and the various tools they tested proved to be slow and inadequate. They determined they ultimately had to find three things: 1) the right tool for the job, 2) the right tech that could handle cluttered and complex data to augment and extend their existing team, and 3) the right partner to provide ongoing guidance and support.
“Collaborating with Superb AI transformed our approach by transitioning us from a non-existent ML data pipeline to a robust, well-defined system for training and testing. As a result, we developed a versatile model adept at handling diverse fields and scenarios, made possible by the exceptional data labeling expertise of Superb AI.”
Chris Norton
Machine Learning Lead, Rowbot
That’s when Rowbot turned to Superb AI’s Data Services. Now, Rowbot can easily send large, complex datasets to Superb AI to review, annotate, and audit while receiving fully labeled data back according to their exact specifications - all within the Superb Label data management solution. This has freed the team to train, evaluate, and refine their models while focusing on improving modeling approaches and their core technology.
Since Rowbot started working with Superb Services, they’ve improved their model’s ability to segment and classify real-world farming data across six categories, significantly boosting overall performance. Outsourcing the labeling process has freed up engineering time and focus on tackling other problems and projects. There have been some unexpected benefits, too - using Superb Services, Rowbot has identified more opportunities for data collection and field testing by uncovering limitations in their model when comparing past performance to current performance.
These labels provide Rowbot with valuable insight into their models, allowing them to identify gaps in their data collection, identify issues, and assess why their models perform the way they do. With this information, they can now make more informed decisions to improve their N-Smart™ technology and, by extension, farming outcomes.
"Without Superb AI, it would have been difficult for us to generate the extensive image annotations within our timeline. We were pleasantly surprised by the amazing teamwork and collaboration we experienced with Superb throughout the project."
Chris Norton
Machine Learning Lead, Rowbot