This example loads the Tiny ImageNet 200 dataset into RAM memory.
In the below example, 90% of the dataset is loaded as training data while 5% is loaded for each validation and testing:
// change ProportionToLoad to a smaller number if you don't have available 6GB of RAM.
ProportionToLoad := 1;
WriteLn('Loading ', Round(ProportionToLoad*100), '% of the Tiny ImageNet 200 dataset into memory.');
CreateVolumesFromImagesFromFolder
(
ImgTrainingVolumes, ImgValidationVolumes, ImgTestVolumes,
{FolderName=}'tiny-imagenet-200/train', {pImageSubFolder=}'images',
{color_encoding=}0{RGB},
{TrainingProp=}0.9*ProportionToLoad,
{ValidationProp=}0.05*ProportionToLoad,
{TestProp=}0.05*ProportionToLoad
);
There is a jupyter notebook that runs this example.