This paper introduces a semi-supervised pipeline that utilizes a generative adversarial network (GAN) to generate unlabeled samples, which are then integrated into a convolutional neural network (CNN) for person re-identification (re-ID). The main contributions of the paper are: 1. **GAN for Unlabeled Data Generation**: The GAN is trained on the original training set to generate new pedestrian images. 2. **Label Smoothing Regularization for Outliers (LSRO)**: A new regularization method is proposed to handle the unlabeled data. LSRO assigns a uniform label distribution to the generated images, preventing the model from overfitting to the generated data. 3. **Evaluation on Person Re-ID Datasets**: The proposed method is evaluated on three large-scale datasets (Market-1501, CUHK03, and DukeMTMC-reID) and shows significant improvements over a baseline CNN, achieving +4.37%, +1.6%, and +2.46% in rank-1 precision, respectively. The paper also discusses related work on GANs, semi-supervised learning, and person re-ID, and provides experimental results to demonstrate the effectiveness of the proposed method. The code for the experiments is available at https://github.com/layumi/Person-reID_GAN.This paper introduces a semi-supervised pipeline that utilizes a generative adversarial network (GAN) to generate unlabeled samples, which are then integrated into a convolutional neural network (CNN) for person re-identification (re-ID). The main contributions of the paper are: 1. **GAN for Unlabeled Data Generation**: The GAN is trained on the original training set to generate new pedestrian images. 2. **Label Smoothing Regularization for Outliers (LSRO)**: A new regularization method is proposed to handle the unlabeled data. LSRO assigns a uniform label distribution to the generated images, preventing the model from overfitting to the generated data. 3. **Evaluation on Person Re-ID Datasets**: The proposed method is evaluated on three large-scale datasets (Market-1501, CUHK03, and DukeMTMC-reID) and shows significant improvements over a baseline CNN, achieving +4.37%, +1.6%, and +2.46% in rank-1 precision, respectively. The paper also discusses related work on GANs, semi-supervised learning, and person re-ID, and provides experimental results to demonstrate the effectiveness of the proposed method. The code for the experiments is available at https://github.com/layumi/Person-reID_GAN.