Real-World Person Re-Identification via Degradation Invariance Learning

Huang, Yukun; Zha, Zheng-Jun; Fu, Xueyang; Hong, Richang; Li, Liang

doi:10.1109/cvpr42600.2020.01409

Cited by 71 publications

(32 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are also works that use GAN to synthesize pedestrian images with different pose, appearance, lighting and resolution for expanding the dataset to improve the generalization ability of the model [9,20,40,66,79,91,135,153,154,165]. Some researchers have also used GAN to learn pedestrian features that are not noise related but identity related to improve the accuracy of feature matching [18,29,41,59]. Based on the characteristics and application scenarios of GAN, we categorize GANbased person Re-ID methods into three categories: imageimage style transfer,data enhancement; and invariant feature learning.…”

Section: Generative Adversarial Networkmentioning

confidence: 99%

“…Several works have enhanced the final feature representation by combining global and local features of pedestrians [13,101,110,120,136,142,147]. Due to its good performance in generating images and feature learning, GAN is widely used for person Re-ID tasks [17,22,29,40,41,72,119,153,154,157,159]. To alleviate the shortage of information in single-frame images, some researchers have used the complementary spatial and temporal cues of video sequences to effectively fuse more information in the video sequences [19,26,36,62,129,132].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep learning-based person re-identification methods: A survey and outlook of recent works

Ming¹,

Zhu²,

Wang³

et al. 2021

Preprint

View full text Add to dashboard Cite

The main studys of person Re-ID surveys have been summarized in recent years.• Deep learning-based person Re-ID methods are classified into five categories according to their characteristic.• The above five categories are subdivided according to their technique types.• This classification is more suitable for researchers to explore these methods from their practical needs.• Furthermore, five possible research directions are analyzed for person Re-ID researchers.

show abstract

Section: Generative Adversarial Networkmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Deep learning-based person re-identification methods: A survey and outlook of recent works

Ming¹,

Zhu²,

Wang³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…It is extensively explored in the literature. Existing methods mainly focus on three categories: designing discriminative hand-crafted descriptors [2], robust distance metric learning [24,50] or deep learning technique [27,39,18,17,16]. For example, Chen et al [5] introduced a cascaded feature suppression mechanism that mines all potential salient features stage-by-stage and integrates these discriminative salience features with the global feature, producing the final pedestrian feature.…”

Section: Related Workmentioning

confidence: 99%

Spatial-Temporal Correlation and Topology Learning for Person Re-Identification in Videos

Liu

Zha

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Video-based person re-identification aims to match pedestrians from video sequences across non-overlapping camera views. The key factor for video person reidentification is to effectively exploit both spatial and temporal clues from video sequences. In this work, we propose a novel Spatial-Temporal Correlation and Topology Learning framework (CTL) to pursue discriminative and robust representation by modeling cross-scale spatial-temporal correlation. Specifically, CTL utilizes a CNN backbone and a key-points estimator to extract semantic local features from human body at multiple granularities as graph nodes. It explores a context-reinforced topology to construct multiscale graphs by considering both global contextual information and physical connections of human body. Moreover, a 3D graph convolution and a cross-scale graph convolution are designed, which facilitate direct cross-spacetime and cross-scale information propagation for capturing hierarchical spatial-temporal dependencies and structural information. By jointly performing the two convolutions, CTL effectively mines comprehensive clues that are complementary with appearance information to enhance representational capacity. Extensive experiments on two video benchmarks have demonstrated the effectiveness of the proposed method and the state-of-the-art performance.

show abstract

“…For distribution alignment approaches, the purpose is to learn domain invariant feature representations. In [6,17,21,28,31,62,63,66,71], generative models such as a generative adversarial network (GAN) are exploited to achieve image-to-image translation from the source domain to the target domain and then use the generated images to train the model. Some other approaches [34] align the feature space by MMD loss.…”

Section: Introductionmentioning

confidence: 99%

MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

As a challenging task, unsupervised person ReID aims to match the same identity with query images which does not require any labeled information. In general, most existing approaches focus on the visual cues only, leaving potentially valuable auxiliary metadata information (e.g., spatio-temporal context) unexplored. In the real world, such metadata is normally available alongside captured images, and thus plays an important role in separating several hard ReID matches. With this motivation in mind, we propose MGH, a novel unsupervised person ReID approach that uses meta information to construct a hypergraph for feature learning and label refinement. In principle, the hypergraph is composed of cameratopology-aware hyperedges, which can model the heterogeneous data correlations across cameras. Taking advantage of label propagation on the hypergraph, the proposed approach is able to effectively refine the ReID results, such as correcting the wrong labels or smoothing the noisy labels. Given the refined results, We further present a memory-based listwise loss to directly optimize the average precision in an approximate manner. Extensive experiments on three benchmarks demonstrate the effectiveness of the proposed approach against the state-of-the-art. CCS CONCEPTS• Information systems → Image search.

show abstract

Real-World Person Re-Identification via Degradation Invariance Learning

Cited by 71 publications

References 8 publications

Deep learning-based person re-identification methods: A survey and outlook of recent works

Deep learning-based person re-identification methods: A survey and outlook of recent works

Spatial-Temporal Correlation and Topology Learning for Person Re-Identification in Videos

MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification

Contact Info

Product

Resources

About