Manually annotated and curated dataset of diverse weed species in maize and sorghum for computer vision

To generate a dataset consisting of high quality images that capture the initial growth dynamics of individual plants of several weed species, a greenhouse experiment was performed at the Moving Fields facility of the LfL in Freising, Germany. This experiment took place from 16.06.2021 - 13.12.2021. The plant species were selected as weed species common to fields of sorghum grown in Germany. A Scanalyzer 3D imaging cabin of the Moving Fields facility was used for generating high-quality images. In this cabin, one RGB camera (Basler piA2400-17gm) is mounted 2.8 m perpendicular above the conveyor band. This camera takes images with 2456 x 2058 pixels, which resulted in a ground sampling distance of ∼ 0.1735 mm per pixel. Each plant in a plot was annotated with bounding boxes using the open souce software CVAT (cvat.ai) throughout the growth, leading to a timeseries of images from sprouting to harvest. The seeds differed in germination rate thus a variable amount of plots were sown per species. This helped mitigating the imbalance issue, but there is still class imbalance in this dataset. The dataset contains 640 plots of 28 weed species (monocots and dicots), 6 varieties of Sorghum and 2 varieties of Maize. Each plot consists of a timeseries of images resulting in > 82.000 images in total.