Unsupervised Domain Adaptive Person Re-id with Local-enhance and Prototype Dictionary Learning

Abstract

The unsupervised domain adaptive person reidentification (re-ID) task has been a challenge because, unlike the general domain adaptive tasks, there is no overlap between the classes of source and target domain data in the person re-ID, which leads to a significant domain gap. State-of-the-art unsupervised re-ID methods train the neural networks using a memory-based contrastive loss. However, performing contrastive learning by treating each unlabeled instance as a class will lead to the problem of class collision, and the updating intensity is inconsistent due to the difference in the number of instances of different categories when updating in the memory bank.To address such problems, we propose Prototype Dictionary Learning for person re-ID which is able to utilize both source domain data and target domain data by one training stage while avoiding the problem of class collision and the problem of updating intensity inconsistency by cluster-level prototype dictionary learning. In order to reduce the interference of domain gap on the model, we propose a local-enhance module to improve the domain adaptation of the model without increasing the number of model parameters. Our experiments on two large datasets demonstrate the effectiveness of the prototype dictionary learning. 71.5% mAP is achieved in the Market-to-Duke task, which is a 2.3% improvement compared to the state-of-theart unsupervised domain adaptive re-ID methods. It achieves 83.9% mAP in the Duke-to-Market task, which improves by 4.4% compared to the state-of-the-art unsupervised adaptive re-ID methods.

ATEAM: Knowledge Integration from Federated Datasets for Vehicle Feature Extraction using Annotation Team of Experts

“…There are several approaches for supervised, semi-supervised, and unsupervised domain adaptation in object recognition [26,25], as well as re-id [10,21,8]. In each case, the solution focuses on generalization between datasets by reducing dataset overfitting, instead of knowledge transfer.…”

“…In each case, the solution focuses on generalization between datasets by reducing dataset overfitting, instead of knowledge transfer. Person re-id approaches use existing re-id datasets to test domain transfer [8,22]. Domain adaptation is common for object detection as well [37].…”

“…While each team can transfer knowledge from {d} k to {d * } k , the labeling is noisy due to domain gaps. Domain gaps existing in cross-dataset evaluation due to differences in dataset distributions [28,8]. Since source and target datasets are drawn independently, there are samples in the target datasets in {d * } k that team members have not generalized to.…”

“…Since each of ATEAM's teams are trained on one subsets of D and used to label a disjoint subset of D without annotations, we must address domain gap [10,21,8,25,26]. Since distributions of datasets are different, some knowledge may be lost and may not map exactly during labeling [28].…”