Measuring Apple Size Distribution From A Near Top-Down Image

The paper presents a method for estimating size information of a bin of apples; we document the machine learning and computer vision techniques applied or developed for solving this task. The system was required to return a statistical distribution of diameter sizes based off visible fruit on the top layer of an apple bin image. A custom data-set was collected before training a Mask R-CNN object detector. Image transformations were used to recover real world dimensions. The presented research was undertaken for further integration into an app where apple growers have the ability to get fast estimations of apple size.