Use Dataset

In LibFewShot, datasets have a fixed format. We read the data according to the datasets in most few-shot learning settings, like miniImageNet and tieredImageNet. Some datasets like Caltech-UCSD Birds 200 can be downloaded from the internet and unzipped for using directly.

If you want to use a new dataset but its data format is different from the above datasets, you need to transform it into the same dataset format.

Dataset format

Like miniImageNet, dataset format should be the same as follows:

dataset_folder/
├── images/
│   ├── images_1.jpg
│   ├── ...
│   └── images_n.jpg
├── train.csv *
├── test.csv *
└── val.csv *

All training, evaluating and testing images should be placed in the images directory, by using train.csv，test.csv and val.csv files to split the dataset, respectively. These three files have similar format, and are organized as follows:

filename    , label
images_m.jpg, class_name_i
...
images_n.jpg, class_name_j

The CSV head contains only two columns, one of which is filename and the other is label. The filename should be a relative path from the images directory. It means that, for an image with absolute path .../dataset_folder/images/images_1.jpg，its filename should be images_1.jpg. In a similar way, for an image with absolute path .../dataset_folder/images/class_name_1/images_1.jpg, its filename should be class_name_1/images_1.jpg.

Configure Dataset

After downloading a dataset and transforming it into the above dataset format, you only need to change the data_root in the configuration file. Notice that LibeFewShot will print the data directory’s name as well as the dataset’s name into the log.