consider both width and height for resizing the input, reported in issue #17

use f1 score as val metric, harder data aug

added license

more data aug

put all contrast operations into one code block, clip bounding boxes to img size

add distractors (lines) in data aug pipeline to simulate lined paper

added early stopping to training loop

fill seg maps up to xmax and ymax, added link in README

typo

init