Camera Style Adaptation for Person Re-identification

By Zhun Zhong, Liang Zheng, Zhedong Zheng, Shaozi Li, Yi Yang. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2017

608 citations recorded by the source corpus.

Being a cross-camera retrieval task, person re-identification suffers from image style variations caused by different cameras. The art implicitly addresses this problem by learning a camera-invariant descriptor subspace. In this paper, we explicitly consider this challenge by introducing camera style (CamStyle) adaptation. CamStyle can serve as a data augmentation approach that smooths the camera style disparities. Specifically, with CycleGAN, labeled training images can be style-transferred to each camera, and, along with the original training samples, form the augmented training set. This method, while increasing data diversity against over-fitting, also incurs a considerable level of noise. In the effort to alleviate the impact of noise, the label smooth regularization (LSR) is adopted. The vanilla version of our method (without LSR) performs reasonably well on few-camera systems in which over-fitting often occurs. With LSR, we demonstrate consistent improvement in all systems regardless of the extent of over-fitting. We also report competitive accuracy compared with the state of the art. Code is available at: https://github.com/zhunzhong07/CamStyle

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/1822ca8db58b0382b0c64f310840f0f875ea02c0

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

How good is my GAN? (2018, background): “Modern GANs generate images realistic enough to improve performance in applications, such as, biomedical imaging [11, 18], person re-identification [58] and image enhancement [55].”
Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification (2017, methodology): “A camera style adaptation method [56] is proposed for re-ID based on CycleGAN.”
Image-to-Image Translation: Methods and Applications (2021, methodology): “I2I methods can also contribute to other visual tasks, such as transfer learning for reinforcement learning [217], image registration [38], domain adaptation [29], [30], [31], person re-identification [218], [219], [220], [221], [222], image segmentation [7], [8], [9], facial geometry reconstruction”
Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (2019, methodology): “Second, as a key influencing factor in person re-ID, camera-style variations [44] might significantly change the appearance of person.”
Distilled Person Re-Identification: Towards a More Scalable System (2019, background): “Unsupervised learning [42, 24, 62, 51, 14, 53, 13, 7, 27, 71], transfer learning [72, 53, 13, 11], small sample learning [63, 2] and active learning [34, 50, 45] are for reducing labelling cost by minimizing the requirement of labelled data or selecting specific data for labelling.”
Cross-Modality Paired-Images Generation for RGB-Infrared Person Re-Identification (2020, methodology): “(Zhong et al. 2018b; 2018a; 2019) translate images to different camera styles with CycleGAN (Zhu et al.”
Style Aggregated Network for Facial Landmark Detection (2018, background): “These mean face images show the very similar face, while they have quite different environments. taking two complementary face images as the input beneﬁts the facial landmark prediction results.”
Multi-Modal Domain Adaptation for Fine-Grained Action Recognition (2020, background): “UDA has been extensively studied for vision tasks including object recognition [14, 15, 32, 33, 51, 55], semantic segmentation [18, 60, 65] and person reidentification [10, 49, 62].”
Self-Similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-Identification (2018, methodology): “ting in [14, 47, 49, 51] to obtain the pre-trained model.”
Adaptive Transfer Network for Cross-Domain Person Re-Identification (2019, methodology): “Recently, a few of UDA approaches tailored for person reidentification [2, 45, 46, 33, 3] have bee proposed upon the domain translation model CycleGAN [48].”
Learning to Reduce Dual-Level Discrepancy for Infrared-Visible Person Re-Identification (2019, methodology): “Person re-identification (re-ID) has recently received increasing attention in the computer vision community [18, 17, 6, 7, 20, 28] because of its great importance in video surveillance.”
Multi-Modal Domain Adaptation for Fine-Grained Action Recognition (2019, background): “UDA has been extensively studied for vision tasks including object recognition [14, 15, 32, 33, 51, 55], semantic segmentation [18, 60, 65] and person reidentification [10, 49, 62].”
Exploit the Unknown Gradually: One-Shot Video-Based Person Re-identification by Stepwise Learning (2018, background): “Recently, a number of deep learning methods are developed [24, 29, 32, 35, 39, 40].”
Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification (2019, methodology): “They use CycleGAN [38] to translate images styles from the source domain to the target domain while keeping image labels unchanged.”
Deep Learning for Person Re-Identification: A Survey and Outlook (2020, methodology): “[33], [75], [98], [116] ; (b) Veriﬁcation Loss [88], [145] and (c) Triplet Loss [13], [48], [84].”
Horizontal Pyramid Matching for Person Re-identification (2018, methodology): “And for CUHK03, we adopt the new training/testing protocol proposed in (Zhong et al. 2017).”
Unsupervised Person Re-Identification via Multi-Label Classification (2020, methodology): “Table 2 also shows that, CamStyle [47] boosts the performance, indicating the importance of data augmentation as discussed in Sec.”
Generalizing a Person Retrieval Model Hetero- and Homogeneously (2018, methodology): “We use ResNet-50 [25] as backbone and follow the training strategy in [35] which ﬁne-tunes on the ImageNet [36] pre-trained model.”