Deep Regression Versus Detection for Counting in Robotic Phenotyping

Work in robotic phenotyping requires computer vision methods that estimate the number of fruit or grains an image. To decide what to use, we compared three for counting grains, each method representative a class approaches from literature. These are two based on density estimation and regression (single multiple column), one object detection. We found when objects image is low, comparable, but as increases, by becomes steadily more accurate than With hundred per image, error count predicted detection-based up 5 times higher using regression-based ones.