Multi-Resolution Outlier Pooling for Sorghum Classification

June 19, 2021
Multiple images from the dataset representing different cultivars. The rows represent different culitvars. The columns represent different captured dates respectively: June 1st, 3rd, 7th, 17th, 19th and 27th, 2017.

This paper introduces the Sorghum-100 dataset, a large dataset of RGB imagery captured by the TERRA-REF field scanner as well as a new and more powerful approach to identifying which plant variety (cultivar) is in a particular image.

The Sorghum-100 dataset was used in the Sorghum-100 Kaggle competition, which explains:

The Sorghum-100 dataset is a curated subset of the RGB imagery captured during the TERRA-REF experiments, labeled by cultivar. This data could be used to develop and assess a variety of plant phenotyping models which seek to answer questions relating to the presence or absence of desirable traits (e.g., "does this plant exhibit signs of water stress?''). In this contest, we focus on the question: ``What cultivar is shown in this image?''