The ACRE Crop-Weed Dataset

For a detailed description of this dataset, based on the Datasheets for Datasets (Gebru, Timnit, et al. "Datasheets for datasets." Communications of the ACM 64.12 (2021): 86-92.), check the ACRE_datasheet.md file.

For what purpose was the dataset created?
The ACRE dataset was created within the scope of the METRICS project to serve as a benchmark for weed detection models in various tasks, including object detection, semantic segmentation, and instance segmentation. The Agri-Food Competition for Robot Evaluation (ACRE) is a benchmarking competition specifically designed for autonomous robots and smart implements, with a primary focus on agricultural activities like weed removal and field navigation. These capabilities play a vital role in facilitating the transition to Digital Agriculture. The ACRE competition, which can be found at https://metricsproject.eu/agri-food, is part of the METRICS project, an EU-funded initiative dedicated to the metrological evaluation and testing of autonomous robots.

What do the instances that comprise the dataset represent?
The instances consist of RGB images depicting both crop and weed plants. The crop category encompasses two species: maize (Zea mays) and beans (Phaseolus vulgaris). On the other hand, the weed category encompasses four species: ryegrass (Lolium perenne), mustard (Sinapis arvensis), matricaria (Matricaria chamomilla), and lamb's quarter (Chenopodium album).

Is there a label or target associated with each instance?
Every image in the dataset is accompanied by an XML file that contains instance segmentation annotations.

What mechanisms or procedures were used to collect the data?
The data collection process involved the use of a four-wheel skid-steering robot that was equipped with a Basler acA2000-50gc RGB camera. The camera was mounted on the robot in such a way that its principal axis was directed perpendicular to the ground. It had a resolution of 2046 x 1080 pixels. The robot was teleoperated and operated at an average speed of 0.2 m/s. To capture the data, the camera's stream was recorded in rosbag format. For this purpose, the camera was connected to a PC running Ubuntu 18.04 and ROS Melodic via an Ethernet interface.