We revisit two popular convolutional neural networks (CNN) in person
re-identification (re-ID), i.e, verification and classification models. The two
models have their respective advantages and limitations due to different loss
functions. In this paper, we shed light on how to combine t