Pose-driven Attention-guided Image Generation for Person Re-Identification

Abstract

Person re-identification (re-ID) concerns the matching of subject images across different camera views in a multi camera surveillance system. One of the major challenges in person re-ID is pose variations across the camera network, which significantly affects the appearance of a person. Existing development data lack adequate pose variations to carry out effective training of person re-ID systems. To solve this issue, in this paper we propose an end-to-end pose-driven attention-guided generative adversarial network, to generate multiple poses of a person. We propose to attentively learn and transfer the subject pose through an attention mechanism. A semantic-consistency loss is proposed to preserve the semantic information of the person during pose transfer. To ensure fine image details are realistic after pose translation, an appearance discriminator is used while a pose discriminator is used to ensure the pose of the transferred images will exactly be the same as the target pose. We show that by incorporating the proposed approach in a person re-identification framework, realistic pose transferred images and state-of-the-art re-identification results can be achieved.

Person Re-Identification Across Multiple Cameras: An Efficient Gated Channel-Spatial Attention Method

“…Many methods [12,[33][34][35][36] focus on designing local-based models to improve feature extraction by dividing the body into different parts and extracting features from each part, which are then fused to create a robust representation. Some methods [11,[37][38][39][40][41][42]use human pose information to accurately detect parts or normalize individuals to gather more information. Somers et al [38] proposed BPBreID, a body part-based Re-ID model for solving occluded person Re-ID problem.…”

“…They also present a deep network, Cross-Correlated Attention Network (CCAN), that utilizes various attention mechanisms to learn robust representations of person images. Khatun et al [41] proposed an end-to-end pose-driven attention-guided generative adversarial network which is designed to generate multiple poses of a person by attentively learning and transferring the subject pose through an attention mechanism. Additionally, an appearance discriminator is utilized to ensure that fine image details remain realistic after pose translation.…”

ReID-DeePNet: A Hybrid Deep Learning System for Person Re-Identification

“…Most of the previously published works directly learn the feature representations from the whole pedestrian image, that contains background clutter. Quite recently, several person ReID deep learning-based systems have suggested learning effective feature representations from the detected pedestrian body to reduce the background clutter and improve the robustness of the person ReID system [6][7][8]. This motivates us to develop an automated image segmentation algorithm to eliminate background noise interference issues and enhance the discriminability of the extracted feature representations, even for an incomplete person, which may contain information that is discriminatory and deserves attention.…”