PAC-GAN: An effective pose augmentation scheme for unsupervised cross-view person re-identification

Abstract

Person re-identification (person Re-Id) aims to retrieve the pedestrian images of a same person that captured by disjoint and non-overlapping cameras. Lots of researchers recently focuse on this hot issue and propose deep learning based methods to enhance the recognition rate in a supervised or unsupervised manner. However, two limitations that cannot be ignored: firstly, compared with other image retrieval benchmarks, the size of existing person Re-Id datasets are far from meeting the requirement, which cannot provide sufficient pedestrian samples for the training of deep model; secondly, the samples in existing datasets do not have sufficient human motions or postures coverage to provide more priori knowledges for learning. In this paper, we introduce a novel unsupervised pose augmentation cross-view person Re-Id scheme called PAC-GAN to overcome these limitations. We firstly present the formal definition of cross-view pose augmentation and then propose the framework of PAC-GAN that is a novel conditional generative adversarial network (CGAN) based approach to improve the performance of unsupervised corss-view person Re-Id. Specifically, The pose generation model in PAC-GAN called CPG-Net is to generate enough quantity of pose-rich samples from original image and skeleton samples. The pose augmentation dataset is produced by combining the synthesized poserich samples with the original samples, which is fed into the corss-view person Re-Id model named Cross-GAN. Besides, we use weight-sharing strategy in the CPG-Net to improve the quality of new generated samples. To the best of our knowledge, we are the first try to enhance the unsupervised cross-view person Re-Id by pose augmentation, and the results of extensive experiments show that the proposed scheme can combat the state-of-the-arts.

Generative Adversarial Networks and Its Applications in Biomedical Informatics

“…Zhang et al (62) used GAN-based model to solve the data shortage problems in person re-identification task. Two view images (cross view images) are generated by a conditional GAN from existing original images and skeleton images.…”

Automatic segmentation, inpainting, and classification of defective patterns on ancient architecture using multiple deep learning algorithms

“…GANs have been extensively researched and applied to images synthesis, 30,31 superresolution, 32,33 caption generation, 34 and image processing, 35 especially in medical aspects. [36][37][38] Current applications of GANs in civil engineering are mostly limited to the augmentation of datasets. Baek et al 39 used GAN to generate construction-equipment images where only the image color was changed to expand the datasets.…”

“…They fight against and learn from each other, eventually reaching a Nash equilibrium, which means the improvement on either side will not lead to an increase in the overall gain and the discriminator can no longer distinguish between generated data and real data. GANs have been extensively researched and applied to images synthesis, 30,31 super‐resolution, 32,33 caption generation, 34 and image processing, 35 especially in medical aspects 36–38 …”

Sequences consistency feature learning for video‐based person re‐identification

“…Zheng et al. [4] concentrate on feature‐level viewpoint transform by turning the matching of two images with different viewpoints to the matching of them with the same viewpoint. Liu et al.…”

“…Zhang et al [3] use the view confusion learning mechanism to learn view-invariant features, which is achieved by adversarial learning between the feature learning and the camera ID guided view classifier. Zheng et al [4] concentrate on feature-level viewpoint transform by turning the matching of two images with different viewpoints to the matching of them with the same viewpoint. Liu et al [5] propose a spatial-temporal correlation and topology learning model, which pursues discriminative and robust features.…”