This dataset is used in the Herbarium 2021 – Half-Earth Challenge and belongs to the technical domain of "large-scale instance-level object recognition". The dataset contains 2.5 million plant sample images collected from multiple large botanical gardens, involving 65,000 categories, covering half of the Earth (Americas, Oceania, etc.). This dataset can be used to train plant recognition algorithms, assist botanists in plant identification, and discover and protect new species.