Case study
Training Models to Identify Citrus
9,390
fruit data points
99.5%
quality
2 passes + QC
on every fruit
The Challenge
Aerobotics forecasts fruit size and yield for growers, packers and marketers. Its models need citrus fruit labelled accurately in real-world imagery, where fruit is easy to miss and hard to tell apart.
What We Did
We labelled citrus fruit to train Aerobotics' identification model. Because accuracy mattered so much, every fruit went through two annotation passes and a final QC check.
The Result
9,390 fruit data points delivered at 99.5% quality.