SnowdenLee

@Ghaleb-alnakhlani As you mentioned you want to train one model for the whole dataset, so the semantics labels should be consistent across the whole dataset not just for one image....