AnnotatedImage currently only holds an image_file_path - and thus an ImageAnnotationDataSet is (also) just a list of paths to image files.
It is assumed that a data set can involve a list of file names which would require more memory to load than what is available (think ImageNet).
It might also be needed to have the memory manager “unload” some earlier loaded files to stay beneath a max memory threshold.
AnnotatedImage currently only holds an image_file_path - and thus an ImageAnnotationDataSet is (also) just a list of paths to image files.
It is assumed that a data set can involve a list of file names which would require more memory to load than what is available (think ImageNet).
It might also be needed to have the memory manager “unload” some earlier loaded files to stay beneath a max memory threshold.