Semi-iNat, short for Semi-Supervised iNaturalist, is a challenging semi-supervised classification dataset with long-tailed distribution categories, fine-grained categories, and domain shifts between labeled and unlabeled data. The dataset contains standard training, validation and test sets. The training set contains annotated images from 810 species, of which about 10% images are annotated.

Citation

@misc{su2021semi_iNat,
title={The Semi-Supervised iNaturalist Challenge at the FGVC8 Workshop},
author={Jong-Chyi Su and Subhransu Maji},
year={2021},
eprint={2106.01364},
archivePrefix={arXiv},
primaryClass={cs.CV}
}