Overview
Orchard Robotics provides farmers with an AI-first approach to precision crop management.
The Problem
Orchard Robotics needed better quality data labels.
To provide farmers with accurate data, Orchard Robotics must accurately count every fruit on every tree in an orchard. Each tree grows roughly 200 fruits, and up to three adjacent trees can be in view in any given image. Early in the season, the fruit might be only five millimeters in diameter, making it just 20 or 30 pixels wide. When the fruit is that small, it is incredibly difficult and tedious to label images with 400-500 detections per image. As a small team of four, Orchard Robotics could not scale these annotations in-house.
Before working with Scale Rapid, Orchard Robotics initially tried acquiring labels using three other major data-labeling services. However, they could not get the consistent quality they needed through these services. They could not provide feedback to the annotators on the quality of the labels, and furthermore, the quality varied dramatically between batches. These other platforms also did not offer ellipses as an annotation type, forcing Orchard Robotics to rely on bounding boxes, a less-than-ideal option when labeling spherical fruit. Orchard Robotics needed a better way to receive high-quality annotations consistently.
The Solution
Scale Rapid provides high-quality annotations with attention to detail.
By switching to Scale Rapid, Orchard Robotics now receives results for their annotation batches in as little as 12 hours, a marked reduction from all the previous services they tried, which took as long as 4-5 days to return results. The quality of their annotations has also improved. Scale Rapid returned high-quality annotations, even for complex, high-resolution images that required serious attention to detail.
Scale Rapid allows users to provide direct feedback to annotators and monitor annotation progress so Orchard Robotics can get reliable, high-quality annotations within a clearly defined timeline. Scale Rapid also automatically controls the quality of the labels and provides an API that allows Orchard Robotics to automate their annotation requests as they continue to grow.
The Result
Orchard Robotics enables precision crop management for farmers.
Orchard Robotics is now able to obtain high-quality annotations through Scale Rapid reliably. Consistent label quality has saved their team a great deal of engineering time and made it easier to train high-performing machine learning models. When relying on inconsistent annotation quality from a different vendor, Orchard Robotics had to manually filter each of their images to eliminate especially low-quality results. Consistent, accurately-labeled data is crucial for detections that require so much precision. For example, if three apples are next to each other, each apple needs to have their own ellipse boundary – any faulty data is likely to compromise the model. Collaborating with Scale Rapid has provided Orchard Robotics with controlled and consistent data quality.
Naturally, higher-quality annotations translate to better results for their customers. Now, Orchard Robotics can more easily and accurately analyze image data, providing farmers with all the data they need to optimize their orchards and maximize crop yield and quality.