CIFAR-FS stands for CIFAR100 few-shots, which is randomly extracted from the CIFAR-100 dataset using the same criteria used to generate miniImageNet. The dataset contains 100 categories, 600 images per category, and a total of 60,000 images. On average, the similarity between categories is high, which is a challenge for the current state of the art. In addition, the image size is 32×32 pixels, and the limited native resolution makes the task more difficult while also allowing for rapid prototyping.

Citation

@inproceedings{ bertinetto2018metalearning, title={Meta-learning with differentiable closed-form solvers}, author={Luca Bertinetto and Joao F. Henriques and Philip Torr and Andrea Vedaldi}, booktitle={International Conference on Learning Representations}, year={2019}, url={https://openreview.net/forum?id=HyxnZh0ct7}, }