Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification

By Weijian Deng, Liang Zheng, Guoliang Kang, Yezhou Yang, Qixiang Ye, Jianbin Jiao. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2017

970 citations recorded by the source corpus.

Person re-identification (re-ID) models trained on one domain often fail to generalize well to another. In our attempt, we present a "learning via translation" framework. In the baseline, we translate the labeled images from source to target domain in an unsupervised manner. We then train re-ID models with the translated images by supervised methods. Yet, being an essential part of this framework, unsupervised image-image translation suffers from the information loss of source-domain labels during translation. Our motivation is two-fold. First, for each image, the discriminative cues contained in its ID label should be maintained after translation. Second, given the fact that two domains have entirely different persons, a translated image should be dissimilar to any of the target IDs. To this end, we propose to preserve two types of unsupervised similarities, 1) self-similarity of an image before and after translation, and 2) domain-dissimilarity of a translated source image and a target image. Both constraints are implemented in the similarity preserving generative adversarial network (SPGAN) which consists of an Siamese network and a CycleGAN. Through domain adaptation experiment, we show that images generated by SPGAN are more suitable for domain adaptation and yield consistent and competitive re-ID accuracy on two large-scale datasets.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/b3dbb682bb8b71723538f433b1dd007ed0b9b244

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

Unsupervised Person Re-Identification via Multi-Label Classification (2020, methodology): “The third category [19, 30, 38, 31, 3, 45, 46, 1, 32, 17, 24, 40] utilizes transfer learning to improve unsupervised person ReID.”
Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-identification (2018, background): “Data synthesis [2, 10] has also been proposed as a solution for addressing limited data, although it suffers from undesirable person appearance distortion and restricted source selection.”
Cluster Contrast for Unsupervised Person Re-Identification (2021, methodology): “The other is the unsupervised domain adaptation person re-ID, which first pre-trains a model on the source labeled dataset, and then fine-tunes the model on the target unlabeled dataset [12,29,48,51,59,67,68].”
Cali-Sketch: Stroke Calibration and Completion for High-Quality Face Image Generation from Poorly-Drawn Sketches (2019, positive): “In recent years, Generative Adversarial Networks (GANs) [23] have been successfully applied in many computer vision tasks to improve realism of generated images, such as domain adaption [24], [25], [26], super-resolution [27], [28], [29], [30], [31], [32].”
Generalizing a Person Retrieval Model Hetero- and Homogeneously (2018, methodology): “Deng et al. [1] learn a similarity preserving generative adversarial network based on CycleGAN [34] to translate images from source domain to target domain.”
Adaptive Transfer Network for Cross-Domain Person Re-Identification (2019, methodology): “Recently, a few cycle generative adversarial Networks (CycleGAN) [48, 1, 7] based UDA approaches [2, 45, 46, 33, 3] are proposed for person re-identification, which focus on learning a generator network that transforms samples in the pixel space from one domain to another.”
A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification (2019, methodology): “In recent years, some transfer learning methods (Hehe et al. 2018; Peng et al. 2016; Deng et al. 2018) are proposed upon the success of deep learning (Dong et al. 2018a; Li, Tang, and Mei 2018).”
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].”
Camera Style Adaptation for Person Re-identification (2017, background): “Recently, GANs have also been applied to image-to-image translation [13, 51, 22], style transfer [8, 14, 6] and cross domain image generation [2, 31, 5].”
Deep Visual Domain Adaptation: A Survey (2018, methodology): “Inspired by cycle GAN and Siamese network, the similarity preserving generative adversarial network (SPGAN) [21] translated the labeled source image to the target domain, preserving self similarity and domain-dissimilarity in an unsupervised manner, and then it trains re-ID models with the translate”
Applications of Generative Adversarial Networks (GANs): An Updated Review (2019, background): “Deng et al. [22] considered the domain adaptation in person re-identification that task aims at searching for images of the same person to the query.”
Unsupervised Domain Adaptive Re-Identification: Theory and Practice (2018, methodology): “Besides, generative models are also applied to domain adaptation in [6, 31].”
Self-Supervised CycleGAN for Object-Preserving Image-to-Image Domain Adaptation (2020, background): “Note that the recent proposed GANs for image-based adaptation, e.g., SPGAN [5], PTGAN [31], and AugGAN [10], are not involved for comparison, due to the strong prior-knowledge used in those approaches.”
Deep Learning for Person Re-Identification: A Survey and Outlook (2020, methodology): “Using the GAN generated images is also a widely used approach in unsupervised domain adaptation Re-ID [118], [119], approximating the target distribution.”
VERI-Wild: A Large Dataset and a New Method for Vehicle Re-Identification in the Wild (2019, background): “GANs have achieved great success in many tasks, such as image generation [15][5] and translation [32][2][3].”
Semantics-Enhanced Adversarial Nets for Text-to-Image Synthesis (2019, methodology): “In several other computer vision tasks, such as unsupersived domain person Re-identification [3, 39, 46] and image-to-image style transfer (CycleGAN) [15], such guidance from groundtruth images has been demonstrated effective in enhancing structural coherence of synthesized images.”
Distilled Person Re-Identification: Towards a More Scalable System (2019, methodology): “We compared with unsupervised ReID methods including unsupervised features LOMO [31], BOW [66] and unsupervised learning models UMDL [42], PTGAN [53], PUL [14], CAMEL [62], SPGAN [13], TJAIDL [51] and HHL [71].”
Style Normalization and Restitution for Generalizable Person Re-Identification (2020, methodology): “UDAbased ReID methods can be roughly divided into three categories: style transfer [3, 44, 26], attribute recognition [43, 49, 33], and target-domain pseudo label estimation [5, 37, 58, 40, 52, 50].”
Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification (2019, methodology): “Finally, it achieves good performance and even surpasses some methods of UDA [4, 6, 26].”