…ime)

2) eval_fscore.py now save metrics to tf.summary as well
3) consolidate tf.summary metrics to some names space to avoid
cluttering

This version achieves 0.879 F1 score on our own street number dataset

2) add adam optimizer option
3) clip bounding box to image boundary
4) add detect_street_number_with_pixellink_model.py to use pixel_link
model for street number detection