This paper presents a novel image dataset with high intrinsic ambiguity and a long-tailed distribution built from the database of Pl@ntNet citizen observatory. It consists of 306146 plant images covering 1081 species. We highlight two particular features of the dataset, inherent to the way the images are acquired and to the intrinsic diversity of plants morphology: (i) the dataset has a strong class imbalance, i.e. a few species account for most of the images, and, (ii) many species are visually similar, rendering identification difficult even for the expert eye. These two characteristics make the present dataset well suited for the evaluation of set-valued classification methods and algorithms. Therefore, we recommend two set-valued evaluation metrics associated with the dataset (macro-average top-k accuracy and macro-average average-k accuracy) and we provide baseline results established by training deep neural networks using the cross-entropy loss. A full description of the dataset as well as baseline experiments can be found in the following publication: "Pl@ntNet-300K: a plant image dataset with high label ambiguity and a long-tailed distribution", Camille Garcin, Alexis Joly, Pierre Bonnet, Antoine Affouard, Jean-Christophe Lombardo, Mathias Chouet, Maximilien Servajean, Titouan Lorieul and Joseph Salmon, in Proc. of Thirty-fifth Conference on Neural Information Processing Systems, Datasets and Benchmarks Track, 2021. Please cite the above reference for any publication using the dataset. Utilities to load the data and train models with pytorch can be found here: https://github.com/plantnet/PlantNet-300K/