A University of León team published the Veridis dataset in Data in Brief on 5 May. It contains 10,080 corn and beet images, including augmented training examples, organised with object-detection annotations and predefined training, validation and test subsets.
The images were collected in fields in León, Spain, using a Robotnik Summit XL equipped with an Intel RealSense D455 camera. A person teleoperated the platform while it recorded daylight observations. The publication therefore supplies evidence for developing perception systems, rather than demonstrating fully autonomous crop inspection.
The authors make the data available through Hugging Face and describe recording-level separation between subsets to reduce leakage from closely related frames. Reported model baselines provide a starting point for comparison, while the limited crops, locations and acquisition conditions constrain wider generalisation.
