Issues with loading very large number of images (e.g., 55000)

[zhuowenzhao]

Labelmaker has latency in displaying the images and data clinic says run out of memory due to loading npz files all at once.
The latency could be due to (1) the sizes of the images are a little higher (1679, 1475) - we may need to resize them to (~512x512) or something like that, (2) I wonder if the image_order vector + maybe the labels in cache could be causing some latency too - since the list is really large.

This might be resolved by storing metadata in splashML or hdf5 files instead of in the memory.