Progressive Unsupervised Person Re-Identification by Tracklet Association With Spatio-Temporal Regularization

Abstract

Existing methods for person re-identification (Re-ID) are mostly based on supervised learning which requires numerous manually labeled samples across all camera views for training. Such a paradigm suffers the scalability issue since in real-world Re-ID application, it is difficult to exhaustively label abundant identities over multiple disjoint camera views. To this end, we propose a progressive deep learning method for unsupervised person Re-ID in the wild by Tracklet Association with Spatio-Temporal Regularization (TASTR). In our approach, we first collect tracklet data within each camera by automatic person detection and tracking. Then, an initial Re-ID model is trained based on within-camera triplet construction for person representation learning. After that, based on the person visual feature and spatio-temporal constraint, we associate cross-camera tracklets to generate cross-camera triplets and update the Re-ID model. Lastly, with the refined Re-ID model, better visual feature of person can be extracted, which further promote the association of cross-camera tracklets. The last two steps are iterated multiple times to progressively upgrade the Re-ID model. To facilitate the study, we have collected a new 4K UHD video dataset named Campus-4K with full frames and full spatio-temporal information. Experimental results show that with the spatio-temporal constraint in the training phase, the proposed approach outperforms the state-of-the-art unsupervised methods by notable margins on DukeMTMC-reID, and achieves competitive performance to fully supervised methods on both DukeMTMC-reID and Campus-4K datasets.

Unsupervised Person Re-Identification with Wireless Positioning under Weak Scene Labeling

“…the model can learn a better feature representation. Xie et al [38] introduce the spatio-temporal constraint to promote the label generation. Lin et al [39] introduce several auxiliary information as additional priors for constraints, including a camera-based term that is useful for distance amendment.…”

“…As shown in the initial training stage of Fig. 3, we use a multi-branch network architecture [34], [38] to train the basic model. The network has a shared feature extraction model F (•) and C independent classifiers.…”

“…Given N video sequences, we extract their features with F (•). Then, we apply the nearest neighbor association (NNA) [34], [38] for cross-camera video sequences association and obtain the pseudo visual labels. Specifically, for each video sequence, NNA first finds cross-camera nearest neighbor on each camera pair based on the cosine similarity of the features.…”

Beyond Triplet Loss: Person Re-Identification With Fine-Grained Difference-Aware Pairwise Loss

“…Inspired by the tremendous success of deep learning, many methods [4], [5], [6] have been introduced to learn deep expressive representations for person ReID and achieved stateof-the-art performance. Typically, most of these methods [7], [8], [4], [9], [5], [10], [11], [12], [6], [13], [14], [15], [16], [17], [18], [19], [20] employ a triplet loss [7], [5], [13] or its combination of a classification loss [10], [11], [12] as the driving force to extract relevant features. Under this generic framework, several approaches have been developed to learn semantically-rich and/or local features, such as the global feature-based approach [14], [15], data augmentation-based approach [6], [13] and striping approach [21], [10].…”

MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification

“…is attached with the captured images. As shown in Figure 1, the metadata could provide auxiliary guidance to unsupervised person ReID [23,24,31,33,40,43,49,51,66,68] Recently, unsupervised person ReID approaches [9,12,60] follow clustering-and-finetune pipeline, which iteratively assigns pseudo labels for data then trains the feature extractor with the pseudo labels. In practice, since the ReID image data is generated from a distributed camera network, the correlation among images is usually diverse, heterogeneous, and complicated.…”