Loading large dataset

[ChungPin]

I'm using Torch to train my network, using image.load to load images.
I'm loading about 80k images and it cost about 8G CPU RAM,
if I need to load larger dataset (ex, 1,000,000 images), is there any function for doing this? I only have 16G RAM

or is there any function like
keras flow_from_directory()
or
pytorch datasets.ImageFolder