Incremental Training - Prioritizing new images

[levipereira]

I'm currently training 100k images. But I add 4k new images every week, make sense add new feature that instead of training the whole dataset again as new , prioritize theses new 4k with a little bit more iteration.
e.g

    train = <path-to-voc>/train.txt
    valid = <path-to-voc>2007_test.txt
    prioritize = <path-to-voc>/train_priorize.txt
    names = date/voc.names
    backup = backup