Fork of brain train to generate backbone models for PEFSL project

• Example scripts for training are available in launch_training.sh,launch_training_resnet12.sh.
• One can evaluate the accuracy of the models in few shot tasks with evaluation.sh.
• to_csv.sh allows to export the results of the evaluation in a csv file.

setup dataset :

working environement :

You can install necessary package using pip :

pip install -r requirements.txt

Setup Datasets

This step is designed to prepare the datasets for later use. In a first place, you must dowload them, and put them into a directory. Once this is done, by running the following command, some meta-data about the dataset (per example number of class, number of samples per class) will be added as a dataset.json in the same folder.

Main available datasets (make sure the name are corrects : ) (check the script for all available datasets)

• miniimagenetimages
• tieredimagenet
• cifar_fs
• imagenet

linux :

python create_dataset_files.py --dataset-path path/to/datasets/folder

adapt the scripts :

change the dataset-path to correspond to your path to the scripts