Person Re-Identification by Optimally Organizing Multiple Similarity Measures

Lin, Lan; Huang, Renjie; Li, Xudong; Zhang, Feng; Ye, Mao

doi:10.1109/access.2017.2771138

Cited by 5 publications

(3 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And some hand-crafted re-id methods are very competitive with their respective strengths particularly on small datasets like CUHK01. Encouragingly, our proposed method behaves robustly and outperforms most competitors except for the rank-1 of MSE-VCM [57].…”

Section: ) Experiments On Cuhk01mentioning

confidence: 79%

See 1 more Smart Citation

Recurrent Models of Visual Co-Attention for Person Re-Identification

et al. 2019

Self Cite

View full text Add to dashboard Cite

Person re-identification (re-id) refers to matching people across disjoint camera views. Most of person re-id methods extract discriminative features from the whole images or fixed regions and develop their metrics. However, these methods ignore that the attention regions with temporal cues in the pedestrian image pair hold discriminative information. In this paper, we propose the recurrent models of visual co-attention that aim to simulate human eye movement, focusing on the sequential concurrent attention (co-attention) regions of the same locations when comparing image pairs. Since reinforcement learning provides a flexible learning strategy for sequential decision-making, it is naturally applied to perform the temporal re-id co-attention learning task. The reward functions are designed to recursively optimize the prediction by rewarding or punishing the learning process. The recurrent models are used to extract information from a sequence of attention regions. Finally, person re-id is performed based on the whole image feature and the features from the recurrent models. Our contributions are: 1) the visual mechanism, which can dynamically locate the optimal co-attention regions to simulate the human re-id process; 2) the design of reward functions in reinforcement learning, which aims to recursively optimize the prediction process; and 3) experimental results, which demonstrate the advantages of our method compared with the state-of-the-art methods. INDEX TERMS Person re-identification, eye movement, reinforcement learning, recurrent neural network, co-attention mechanism.

show abstract

Section: ) Experiments On Cuhk01mentioning

confidence: 79%

“…The cumulative matching scores (%) at rank-1, 5, 10 and 20, as well as mAP (%) are listed. four methods, including LOMO + XQDA [1], MLAPG [33], GOG [2] and MSE-VCM [57]. First, we choose one split from the experiments on labeled CUHK03 dataset.…”

Section: ) Verification Of Images With Occlusionmentioning

confidence: 99%

Recurrent Models of Visual Co-Attention for Person Re-Identification

et al. 2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…On VIPeR dataset, our method achieves accuracy of 64.9% at rank1, which is higher than the rate of the other methods. In particular, compared with SLME [35], MSE-VCM [39] based on multiple metrics and MPML [38], EquiDML(Fusion) [40] based on multiple feature fusion, the performance of our method is about 14.6%, 13.8%, 14.9% and 13.5% higher at rank-1, respectively. Besides, our method has the best re-ID rate of rank-1 on the three datasets, which is an important indicator of the performance of an algorithm.…”

Section: Performance Analysismentioning

confidence: 92%

A Weighted Center Graph Fusion Method for Person Re-Identification

Geng

Guo

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Feature fusion is widely used in person re-identification (re-ID) and has been proven effective. However, it is difficult to know which features are effective to identify a specific person and how to fuse features to explore complementary information and apply the advantages of each feature. Motivated by these problems, this paper proposes a new method of person re-ID to fuse the recognition results of multiple features at the rank level. Three innovations are included in this method: first, multiple metric spaces are constructed based on the correlation of different features to generate multiple rank results; second, the most similar candidates in each corresponding rank list is converted into a graph structure by our proposed weighted center graph (WCG), and we use an adaptive value K to automatically seek the most similar images of each query, thus improving the accuracy of candidate targets. Finally, to evaluate the effect of each WCG, a discriminative power coefficient is designed and used to assign a proper coefficient for each WCG according to the discriminative power of corresponding features. The result can be obtained by re-ranking the fused WCG. The extensive experiments on five datasets demonstrate the matching rate of our proposed method by comparing with several state-of-the-art methods. Our code is available at https://github.com/gengshuze/WCG.git. INDEX TERMS Person re-identification, weighted center-graph, multiple metrics, fusion graph.

show abstract