Being a cross-camera retrieval task, person re-identification suffers from
image style variations caused by different cameras. The art implicitly
addresses this problem by learning a camera-invariant descriptor subspace. In
this paper, we explicitly consider this challenge by introduci