Super-resolution-based part collaboration network for vehicle re-identification.

Vehicle re-identification (ReID) is to find a particular vehicle across different surveillant cameras in city roads. There are two main challenges in vehicle ReID. In one case, pictures of the same vehicle from different camera views often look different, which is called "intra-instance difference" in this paper. In another case, different vehicles of the same model and color often look similar from the same camera view, which is called "inter-instance similarity" in this paper. To deal with these challenges, we design a super-resolution-based part collaboration network (SPCN) for vehicle ReID, which aims to improve the performance of ReID system by paying attention to local features with strong differentiation. First, a fine-grained object detection method is used to find the discriminative local parts in vehicle images, and the image qualities of these local parts are enhanced by a super-resolution model with long skip connections. Then, a part collaboration mechanism is designed to realize the weighted fusion of local features, since different local features show different importance in the process of vehicle ReID task. Finally, the global features are fused with the local features to produce the identification results. Extensive experiments on two public data sets show the effectiveness of our method, which introduces less background noise and achieves state-of-the-art results. [ABSTRACT FROM AUTHOR]