The authors introduce CWFID: A Crop/Weed Field Image Dataset as a standardized dataset for addressing tasks such as distinguishing between crops and weeds, conducting single plant phenotyping, and other computer vision challenges in precision agriculture. These images were captured in an organic carrot farm during the early true leaf growth stage of the carrot plants, utilizing the autonomous field robot Bonirob. The dataset encompasses scenarios involving both intra- and inter-row weeds, where weeds and crops share similar size and grow in close proximity.
Each image in the dataset is accompanied by a ground truth vegetation segmentation mask, and the authors have manually annotated the plant type (crop or weed) for further reference. They also present preliminary findings for solving the crop/weed classification phenotyping problem and introduce evaluation methodologies to facilitate comparisons between various approaches.