Person re-identification (ReID) goals to discover individuals across multiple non-overlapping cameras. The challenge of obtaining comprehensive datasets has driven the necessity for data augmentation, with generative adversarial networks (GANs) emerging as a promising solution.

Techniques like GAN and its variant, deep convolutional generative adversarial networks (DCGAN), have been used to generate human images for data augmentation. The Camera style (CamStyle) using CycleGAN addresses the difficulty of various camera styles, while the pose-normalized GAN (PNGAN) focuses on capturing different pedestrian postures. The first challenge is matching individuals across various camera styles. GAN-based methods often produce unlabeled images, and while some techniques reduce camera style differences, they’ll introduce noise and redundancy. The range in pedestrian postures across cameras also presents a challenge.

A research team from China published a brand new paper to beat the challenges cited above. The authors introduced an improved CycleGAN for ReID data augmentation. Their method integrates a pose constraint sub-network, ensuring consistency in posture while learning camera style and identity. In addition they employ the Multi-pseudo regularized label (MpRL) for semi-supervised learning, allowing for dynamic label weight project. Preliminary results indicate superior performance on multiple ReID datasets.

The whole system comprises two generator networks, two discriminator networks, and two semantic segmentation networks. These segmentation networks are termed pose constraint networks and are instrumental in ensuring consistency in pedestrian postures across different images. Within the improved CycleGAN, first, a generator is tasked with creating fake images, and the discriminator assesses the authenticity of those pictures. Through a continuous iterative process, the generated images are progressively refined to resemble real images closely. A major feature of this approach is the pose constraint loss, which ensures the posture of 1 domain (X) aligns with the opposite domain (Y). This loss is computed by measuring the pixel distance between the fake and real images.

Moreover, the CycleGAN utilizes cyclic consistency to map generated images back to their source domain, ensuring the integrity of transformations. To enhance the performance of the improved CycleGAN, a training strategy has been outlined. This strategy involves using image annotation tools, pre-training specific sub-networks, and repeatedly optimizing the overall loss function.

Lastly, the paper introduces the Multi-pseudo regularized label (MpRL) method, designed to assign labels to generated images more effectively than traditional semi-supervised learning techniques. The MpRL offers various weights to different training classes, allowing for more refined and accurate labeling of generated images and improving pedestrian re-identification results. This method contrasts with the LSRO strategy, which tends to offer uniform weights to all training classes, often leading to less accurate predictions.

To judge the efficiency of the proposed method, the authors tested on three-person re-identification (ReID) datasets: Market-1501, DukeMTMC-reID, and CUHK03-NP. These datasets confront challenges like color differences between cameras and data imbalance. Rank-n and mAP were the first evaluation metrics used. The experiment was inbuilt Python3 with PyTorch on a strong Linux server. Initially, an improved CycleGAN network was trained for camera discrepancies, followed by the ReID network. For validation, the authors conducted an ablation study. The improved CycleGAN yielded higher rank-1 and mAP scores than the usual CycleGAN. The most effective hyperparameters for the CycleGAN were determined experimentally. Comparisons between the LSRO and MpRL methods revealed that MpRL was superior. Incorporating various popular loss functions with MpRL had various effects on performance. The outcomes established that using the improved CycleGAN with the MpRL method outperformed conventional data augmentation techniques, effectively bridging camera style differences and enhancing re-identification accuracy. Comparing the proposed method against other state-of-the-art methods further corroborated the prevalence of their approach.

To conclude, the research team introduced a sophisticated CycleGAN for person re-identification, embedding a pose constraint sub-network to diminish camera style variances. Pose constraint losses maintain posture consistency during identity learning. MpRL is used for label allocation, enhancing re-identification precision. Evaluations on three ReID datasets confirm their method’s efficacy. Future efforts will deal with domain variances to optimize the model for real-world scenarios.

Try the Paper. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to affix our 30k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the most recent AI research news, cool AI projects, and more.

For those who like our work, you’ll love our newsletter..