Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in Vitro

By Zhedong Zheng, Liang Zheng, Yi Yang. 2017

2,052 citations recorded by the source corpus.

The main contribution of this paper is a simple semisupervised pipeline that only uses the original training set without collecting extra data. It is challenging in 1) how to obtain more training data only from the training set and 2) how to use the newly generated data. In this work, the generative adversarial network (GAN) is used to generate unlabeled samples. We propose the label smoothing regularization for outliers (LSRO). This method assigns a uniform label distribution to the unlabeled images, which regularizes the supervised model and improves the baseline. We verify the proposed method on a practical problem: person re-identification (re-ID). This task aims to retrieve a query person from other cameras. We adopt the deep convolutional generative adversarial network (DCGAN) for sample generation, and a baseline convolutional neural network (CNN) for representation learning. Experiments show that adding the GAN-generated data effectively improves the discriminative ability of learned CNN embeddings. On three large-scale datasets, Market- 1501, CUHK03 and DukeMTMC-reID, we obtain +4.37%, +1.6% and +2.46% improvement in rank-1 precision over the baseline CNN, respectively. We additionally apply the proposed method to fine-grained bird recognition and achieve a +0.6% improvement over a strong baseline. The code is available at https://github.com/layumi/ Person-reID_GAN.

How researchers see this paper

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

Citations in the corpus

Continual semi-supervised learning through contrastive interpolation consistency (2021, background): “Recently, several works investigated the refinement of such regularization through adversarial training, producing either more challenging perturbations [29] or additional unsupervised samples for regularization purposes [30].”
Deep Mutual Learning (2017, background): “In contrast many recently proposed deep re-id networks such as those in [31, 23, 34] have complicated and specially designed architectures to handle the drastic pose-changes and body-part mis-alignment when matching people across camera views.”
In Defense of the Triplet Loss for Person Re-Identification (2017, background): “In many aspects person ReID is similar to image retrieval, where significant progress has been made and where deep learning has recently introduced a lot of changes.”
Beyond Face Rotation: Global and Local Perception GAN for Photorealistic and Identity Preserving Frontal View Synthesis (2017, background): “Recently, modiﬁed GAN architectures, conditional GAN [21] in particular, have been successfully applied to vision tasks like image inpainting [24], super-resolution [19], style transfer [20], face attribute manipulation [29] and even data augmentation for boosting classiﬁcation models [30,43].”
Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification (2019, background): “DukeMTMC-reID DukeMTMC-reID [34] is a subset of pedestrian re-identification dataset DukeMTMC [17].”
Camera Style Adaptation for Person Re-identification (2017, methodology): “Examples generated by our method and DCGAN in [47].”
A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification (2019, methodology): “We achieves top-1 accuracy of 61.9% on Market-1501 (Zheng et al. 2015) and 55.1% on MARS (Zheng et al. 2016).”
A Strong Baseline and Batch Normalization Neck for Deep Person Re-Identification (2019, methodology): “On the basis of [29], PTGAN [30] and CamStyle [31] were proposed to bridge domain and camera gaps for person ReID, respectively.”
Pose-Guided Feature Alignment for Occluded Person Re-Identification (2019, methodology): “DukeMTMC-reID [46, 28] contains 1,404 identities, 16,522 training images, 2,228 queries, and 17,661 gallery images.”
One Venue, Two Conferences: The Separation of Chinese and American Citation Networks (2022, methodology): “Furthermore, researchers in China have continued to use the Duke dataset even after its removal [17, 7, 19].”
Pedestrian Alignment Network for Large-scale Person Re-Identification (2017, methodology): “In this section, we report the results on three large-scale datasets: Market-1501 [Zheng et al., 2015], CUHK03 [Li et al., 2014] and DukeMTMC-reID [Zheng et al., 2017b].”
VERI-Wild: A Large Dataset and a New Method for Vehicle Re-Identification in the Wild (2019, methodology): “Recent ReID methods also explore GAN both in vehicle and person ReID fields [30][22][13][12].”
Human Semantic Parsing for Person Re-identification (2018, methodology): “33% in rank-1 and DukeMTMC-reID [50] by 8.”
Perceive Where to Focus: Learning Visibility-Aware Part-Level Features for Partial Person Re-Identification (2019, methodology): “On two holistic re-ID datasets, Market1501 and DukeMTMC-reID, there do exist some images which contain either partial pedestrian or oversized bounding boxes.”
Generative Adversarial Network Based Acoustic Scene Training Set Augmentation and Selection Using SVM Hyper-Plane (2017, methodology): “In the GAN for 2-D images, the convolutional layer of DCGAN is generally used [4-6, 9] for 2-D matrix, but in this work, we used FCNN to process the feature vector (800 x 1 [dim.”