DeepReID: Deep Filter Pairing Neural Network for Person Re-Identification

这篇论文没有代码，只是简单地了解一下，毕竟是首次将深度学习应用到行人再识别任务中的论文。

Abstract

      Person re-identification is to match pedestrian images from disjoint camera views detected by pedestrian detectors. Challenges are presented in the form of complex variations of lightings, poses, viewpoints, blurring effects, image resolutions, camera settings, occlusions and background clutter across camera views. In addition, misalignment introduced by the pedestrian detector will affect most existing person re-identification methods that use manually cropped pedestrian images and assume perfect detection.

       行人再识别是为了匹配从行人探测器探测到的不相交的摄像机视图中的行人图像。 挑战以整个相机视图中光照，姿势，视点，模糊效果，图像的分辨率，相机设置，遮挡和杂乱背景复杂变化的形式呈现。 此外，由行人探测器引入的未对齐将影响大多数现有的人重新识别的方法使用手动裁剪行人图像并假设完美检测。

        In this paper, we propose a novel filter pairing neural network (FPNN) to jointly handle misalignment, photometric and geometric transforms, occlusions and background clutter. All the key components are jointly optimized to maximize the strength of each component when cooperating with others. In contrast to existing works that use handcrafted features, our method automatically learns features optimal for the re-identification task from data. The learned filter pairs encode photometric transforms. Its deep architecture makes it possible to model a mixture of complex photometric and geometric transforms. We build the largest benchmark re-id dataset with 13,164 images of 1,360 pedestrians. Unlike existing datasets, which only provide manually cropped pedestrian images, our dataset provides automatically detected bounding boxes for evaluation close to practical applications. Our neural network significantly outperforms state-of-the-art methods on this dataset.

       在本文中，我们提出了一种新颖的滤波器配对神经网络（FPNN），以共同处理未对准，光照和几何变换，遮挡和杂乱背景。 所有关键部件都经过联合优化，以便在与其他部件合作时最大化每个部件的强度。 与使用手工设计特征的现有作品相比，我们的方法会自动从数据中学习重新识别任务的特征。 学习的滤波器对编码光度变换。 其深层结构使得模拟复杂光度和几何变换的混合成为可能。 我们建立了最大的基准重识别数据集，包含1,360个行人的13,164张图像。 与仅提供手动裁剪的行人图像的现有数据集不同，我们的数据集提供自动检测的边界框，以便在实际应用中进行评估。 我们的神经网络明显优于该数据集上最先进的方法。

1. Introduction

        The purpose of person re-identification is to match pedestrians observed in non-overlapping camera views with visual features [13, 9, 35, 1, 7, 3, 51, 20, 19, 16, 29, 26, 24,48, 2, 41]. It has important applications in video surveillance, such as cross-camera tracking [42], multi-camera event detection [27], and pedestrian retrieval [27]. This problem is extremely challenging because it is difficult to match the visual features of pedestrians captured in different camera views due to the large variations of lightings, poses, viewpoints, image resolutions, photometric settings of cameras, and cluttered backgrounds. Some examples are shown in Figure 1.

        行人再识别的目的是利用视觉特征[13, 9, 35, 1, 7, 3, 51, 20, 19, 16, 29, 26, 24,48, 2, 41]匹配在非重叠视图中观察到的行人。它在视频监督中有重要的应用，比如交叉视图追踪[42]，多视图事件监测[27]，以及行人检索[27]。问题是极具挑战的因为很难匹配在不同的视图中捕获的行人特征，由于光照、姿态、视角、图像分辨率、相机的光度设置不同以及杂乱的背景。一些示例展示在图1中。

                          图1 行人再识别中不同摄像机视图中观察到的行人图像样本。相邻的图像有相同的身份。

        The typical pipeline of a person re-identification system is shown in Figure 2. In practice, it should start with automatic pedestrian detection, which is an essential step for extracting pedestrians from long-hour recorded videos.Given a pedestrian detection bounding box, manually designed features are used to characterize the image region in all the existing works, although they may be suboptimal for the task of person re-identification. Image regions of the same person undergo photometric transforms due to the change of lighting conditions and camera settings. Their geometric transforms are caused by misalignment and the change of viewpoints and poses. Such transforms could be normalized by learning mapping functions [33, 34] or similarity metrics [16, 51]. It is also supposed to be robust to occlusions and background clutter. All the existing works optimize each module in the pipeline either separately or sequentially. If useful information is lost in previous steps,
it cannot be recovered later. Establishing automatic interaction among these components in the training process is crucial for the overall system performance.

       典型的行人再识别系统流程如图2所示。实际中，它应该以自动检测行人开始，这是从长时间记录的视频中提取行人至关重要的一步。给定一行人检测边界框，在现有的工作中，手工设计的特征被用于表征图像区域，虽然它们对于行人再识别任务来说也许是次优的。由于照明条件和相机设置的改变，