Unsupervised Salience Learning for Person Re-identification

By Rui Zhao, Wanli Ouyang, Xiaogang Wang. 2013 IEEE Conference on Computer Vision and Pattern Recognition, 2013

1,120 citations recorded by the source corpus.

Human eyes can recognize person identities based on some small salient regions. However, such valuable salient information is often hidden when computing similarities of images with existing approaches. Moreover, many existing approaches learn discriminative features and handle drastic viewpoint change in a supervised way and require labeling new training data for a different pair of camera views. In this paper, we propose a novel perspective for person re-identification based on unsupervised salience learning. Distinctive features are extracted without requiring identity labels in the training procedure. First, we apply adjacency constrained patch matching to build dense correspondence between image pairs, which shows effectiveness in handling misalignment caused by large viewpoint and pose variations. Second, we learn human salience in an unsupervised manner. To improve the performance of person re-identification, human salience is incorporated in patch matching to find reliable and discriminative matched patches. The effectiveness of our approach is validated on the widely used VIPeR dataset and ETHZ dataset.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/cf78a78cd69e9ddb89a951be81c4972ffbf252bd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

A Bi-Directional Message Passing Model for Salient Object Detection (2018, methodology): “It can be taken as a pre-processing step in many computer vision tasks, such as scene classification [25], visual tracking [3, 21], person re-identification [40] and image retrieval [5].”
Visual saliency based on multiscale deep features (2015, background): “Recently, visual saliency has also been increasingly used by visual recognition tasks [32], such as image classification [36] and person re-identification [39].”
Pose-Driven Deep Convolutional Model for Person Re-identification (2017, background): “them [11, 6, 33, 29, 60, 51, 64]; (b) closing the gap between a person’s different features by learning a discriminative distance metric [32, 9, 17, 36, 55, 23, 54, 30, 26, 65, 50, 3, 27, 4, 39, 28, 10, 37, 59].”
Efficient PSD Constrained Asymmetric Metric Learning for Person Re-Identification (2015, neutral): “27 eSDC [31] 19.”
Unsupervised Cross-Dataset Transfer Learning for Person Re-identification (2016, methodology): “Some state-of-the-art unsupervised Re-ID methods are selected for comparison, including the hand-crafted-feature-based method SDALF [6], the saliency-learning-based eSDC [38], the graphical-model-based GTS [34] and the sparserepresentation-classification-based ISR [24].”
DeepReID: Deep Filter Pairing Neural Network for Person Re-identification (2014, methodology): “In order to increase the discriminative power, patch-based local features have been used [13, 9, 1, 25, 26, 48, 47, 49, 24].”
Deep feature learning with relative distance comparison for person re-identification (2015, methodology): “Figure 5 shows the curves of the various models, and Comparison on VIPeR dataset Using the VIPeR dataset, we compared our method with such state-of-the-art methods as MtMCML [30], LMLF [13], SDALF [5], eBiCov [31], eSDC [32], PRDC [9], aPRDC [33], PCCA [14], KISSME [34], LF [35] and SalMatch [36].”
Salient Object Detection in the Deep Learning Era: An In-Depth Survey (2019, neutral): “In computer vision, sampled applications of SOD include image understanding [1], [2], image cap-tioning [3]–[5], object detection [6], [7], un-supervised video object segmentation [8], [9], semantic segmentation [10]– [12], person re-identiﬁcation [13], [14], etc .”
Deep Saliency with Encoded Low Level Distance Map and High Level Features (2016, background): “This topic has received a great deal of attention in computer vision and cognitive science because of its wide range of applications such as content-aware image cropping [22] and resizing [3], video summarization [24], object detection [20], and person re-identiﬁcation [31].”
Deep Learning for Person Re-Identification: A Survey and Outlook (2020, background): “To handle the outlying regions or background clutter within the person image, pose estimation cues [16], [17] or attention cues [57], [84], [227] are exploited.”
An improved deep learning architecture for person re-identification (2015, methodology): “When using metric learning methods and Euclidean distance, dense color histograms and dense SIFT are used [30].”
Performance Measures and a Data Set for Multi-target, Multi-camera Tracking (2016, background): “Discriminative power is improved by saliency information [33, 34] or learning features specific to body parts [6, 18, 20–23, 27], either in the image [35–37] or back-projected onto an articulated [38, 39] or monolithic [40] 3D body model.”
Saliency detection by multi-context deep learning (2015, background): “Saliency information has also been exploited in high-level vision tasks, such as object detection [37], and person re-identification [62, 61].”
Deep Metric Learning for Person Re-identification (2014, background): “The most popular features include HSV color histogram [8], [5], LAB color histogram [9], SIFT [9], LBP histogram [5], Gabor features [5] and their fusion.”
Person Re-identification: Past, Present and Future (2016, methodology): “The architecture designed by Li et al. [16] is different in that a patch matching layer is added which multiplies the convolution responses of two images in different horizontal stripes, similar to ACS [30] in spirit.”