mlx.data.datasets.load_imagenet

mlx.data.datasets.load_imagenet(root=None, split='train', quiet=False, validate_download=True, tar_index_threads=None)

Load the ImageNet dataset from the downloaded archives.

ImageNet cannot be automatically downloaded so you have to manually download it from http://image-net.org/ . You need the split you want to load and the devkit for tasks 1 and 2.

Parameters:

• root (Path or str, optional) – The directory to load the data from. If none is given then ~/.cache/mlx.data/imagenet is used. However, if the data is not there it cannot be downloaded automatically.

• split (str) – The split to use. It must be either ‘train’ or ‘val’.

• quiet (bool) – If true do not show progress bars.

• validate_download (bool) – If true validate the download if it isn’t already validated.

• tar_index_threads (int, optional) – How many threads to use to index the nested tar file for the imagenet training set. This is not used for the validation set or if the tar file is extracted.