SBSGAN: Suppression of Inter-Domain Background Shift for Person Re-Identification

Cross-domain person re-identification (re-ID) is challenging due to the bias between training and testing domains. We observe that if backgrounds in the training and testing datasets are very different, it dramatically introduces difficulties to extract robust pedestrian features, and thus compromises the cross-domain person re-ID performance. In this paper, we formulate such problems as a background shift problem. A Suppression of Background Shift Generative Adversarial Network (SBSGAN) is proposed to generate images with suppressed backgrounds. Unlike simply removing backgrounds using binary masks, SBSGAN allows the generator to decide whether pixels should be preserved or suppressed to reduce segmentation errors caused by noisy foreground masks. Additionally, we take ID-related cues, such as vehicles and companions into consideration. With high-quality generated images, a Densely Associated 2-Stream (DA-2S) network is introduced with Inter Stream Densely Connection (ISDC) modules to strengthen the complementarity of the generated data and ID-related cues. The experiments show that the proposed method achieves competitive performance on three re-ID datasets, i.e., Market-1501, DukeMTMC-reID, and CUHK03, under the cross-domain person re-ID scenario.

1. Introduction

The task of person re-identification (re-ID) is to match the identities of a person under non-overlapped camera views [10], [39], [18], [41], [24]. Most existing methods assume that the training and testing images are captured from the same scenario. However, this assumption is not guaranteed in many applications. For instance, person images captured from two different campuses have distinct illumination condition and background (BG) (e.g., Market-1501 [38] and DukeMTMC-reID [29], [41] datasets). In this situation, the bias between data distributions on two domains becomes large. Directly training a classifier from one dataset (i.e., source domain) often produces a degraded performance when testing is conducted on another dataset (i.e., target domain). Therefore, it is important to investigate solutions for such a cross-domain issue. For person re-ID, the domain adaption solutions have drawn attention in recent years [4], [7], [35], [36], [42]. Figure 1.

Comparison between different input images for cross-domain person re-ID. Images from Market-1501 and DukeMTMC-reID show distinct BG shift. Images generated by SPGAN [7] and PTGAN [36] do not suppress the BG noise, and have the BG shift problem. The hard-mask solution, i.e., JPP-Net [25] damages the FG. Our SBSGAN takes all the impacts into consideration.