Consider using a smaller sample dataset

[ioalexei]

Walking through the ReadMe examples using the default dataset provided (Three Rivers) results in fetching 3,500 files from Mapillary, using 45Gb storage. On my reasonably fast connection (150Mbps down) it took around 3 hours to download the images from Mapillary, plus another hour to run assign_images.

Consider using a smaller sample dataset (or at providing a smaller dataset as a secondary option) to make it quicker and simper to get started with testing.