A convenience script to download all data sets for DeepOBS and preprocess them so they are ready to be used with DeepOBS.
Currently there is no data downloading and preprocessing mechanic implemented for ImageNet. Downloading the ImageNet data set requires an account and can take a lot of time to download. Additionally, it requires quite a large amount of memory. The best way currently is to download and preprocess the ImageNet data set separately if needed and move it into the DeepOBS data folder.
The file will create a set of folders of the following structure:
DeepOBS expects a structure like this, so if you already have (most of the) the data sets already, you still need to bring it into this order.
usage: deepobs_prepare_data.sh [--data_dir=DIR] [--skip SKIP] [--only ONLY]
|Path where the data sets should be saved. Defaults to the current folder.
|Defines which data sets should be skipped. Argument needs to be one of the following
tolstoi. You can use the
--skip argument multiple times.
|Specify if only a single data set should be downloaded. Argument needs to be one of the following
tolstoi. This overwrites the
--skip argument and should can only be used once.