The OPPD: Open Plant Phenotyping Database is a publicly available dataset consisting of 7590 RGB images featuring 47 plant species. Each species undergoes cultivation under three distinct growth conditions, ensuring a diverse visual appearance. These images are gathered at Aarhus University, Research Centre Flakkebjerg, Denmark, utilizing a custom data acquisition platform that offers images at a ground resolution of approximately 6.6 pixels per millimeter. Note, that the actual number of plant species is 52

The imaging setup consisted of a 12.3 Megapixel global shutter camera (Flir GS3-U3-123S6C) paired with a 38mm lens at F/4 (Schneider Kreuznach Xenon-Topaz XN 2,0/38-0901). To ensure well-illuminated, high-resolution images, a Xenon ring flash (Paul C. Buff, AlienBee ABR800) was used, with the lens positioned at the center of the ring. The imaging system was mounted at a height of 1.7 meters above the seedbeds, providing a ground field of view measuring 0.62×0.45 meters and resulting in a ground resolution of approximately 6.6 pixels per millimeter.

Recording was conducted by moving the imaging system over each of the polystyrene boxes 1–3 times daily, depending on the season of the trial. The tracking of plants in each trial extended from seedling emergence to the 6 to 8 leaf stage, covering a developmental period spanning from 36 to 60 days. These stages are relevant for weed control in field settings. Data collection was carried out across four trial seasons: 2017 spring, 2017 autumn, 2018 summer, and 2019 summer.

Individual plant annotations within each image were carried out using the RoboWeedMaPS online tool (I·GIS A/S in Risskov, Denmark). This tool offered a machine learning-assisted annotation framework for creating bounding boxes and annotating plant species. Initially, the RoboWeedMaPS annotation tool provided region proposals and made preliminary classifications, distinguishing between monocotyledon and dicotyledon plants within a given image. Consequently, the manual annotator’s task primarily involved validating these proposals and rectifying any potential errors.

In order to provide a broad spectrum of plant appearances, the authors cultivated plant species in polystyrene boxes (0.40×0.40×0.15 m) under three different controlled growth conditions, partly inducing stress. The different growth conditions were as follows: G1 involved a potting mixture consisting of sandy loam soil, sand, and peat (2:1:1 w/w) with all necessary micro- and macro-nutrients, and optimum soil moisture; G2 utilized sandy loam soil with optimum fertilizer supply and sub-optimum watering; and G3 used sandy soil with low nutrient content and optimum soil moisture. Initially, the growth boxes were watered to field capacity, and seeds were placed at the soil surface and covered with approximately 1 cm of soil. Subsequently, the boxes were placed on outdoor tables and watered from below three times a day, with occasional light showers applied on top to maintain surface moisture. After seedling emergence, the authors thinned the plants to minimize overlapping plant leaves. For the sandy loam soil (condition G2), drought stress was induced in the plants by stopping watering when the seedlings had 2 true leaves. The authors prevented natural rain using an automated rain cover that was activated only when nearby rain sensors detected rain. All seeds used in the experiments were provided by the seed bank at Research Centre Flakkebjerg, Aarhus University.

The dataset comes complete with annotations using the EPPO encoding system for plant species, bounding box annotations for plant detection and individual extraction, information about the growth conditions, and time elapsed since seeding. Information about the trial id, box id, date and time since seeding are represented as image’s filename. EPPO code and growth condition are applied as tags on each image.

Here is how EPPO codes for plants generally work: codes consist of 5 letters = 3 (genus) + 2 (species). For example, Alopecurus myosuroides -> ALOMY

More information on that here