DeepWeeds is the first large-scale public multi-class weed image dataset, allowing the development of powerful classification methods to make robotic weed control feasible. The dataset contains 17,509 labeled images of eight nationally important weed species in Australia.