Learning to detect partially labeled people

摘要

Deployed vision systems often encounter image variations poorly represented in their training data. While observing their environment, such vision systems obtain unlabeled data that could be used to compensate for incomplete training. In order to exploit