Learning to rank in person re-identification with metric ensembles

By Sakrapee Paisitkriangkrai, Chunhua Shen, Anton van den Hengel. Computer Vision and Pattern Recognition, 2015

481 citations recorded by the source corpus.

We propose an effective structured learning based approach to the problem of person re-identification which outperforms the current state-of-the-art on most benchmark data sets evaluated. Our framework is built on the basis of multiple low-level hand-crafted and high-level visual features. We then formulate two optimization algorithms, which directly optimize evaluation measures commonly used in person re-identification, also known as the Cumulative Matching Characteristic (CMC) curve. Our new approach is practical to many real-world surveillance applications as the re-identification performance can be concentrated in the range of most practical importance. The combination of these factors leads to a person re-identification system which outperforms most existing algorithms. More importantly, we advance state-of-the-art results on person re-identification by improving the rank-1 recognition rates from 40% to 50% on the iLIDS benchmark, 16% to 18% on the PRID2011 benchmark, 43% to 46% on the VIPeR benchmark, 34% to 53% on the CUHK01 benchmark and 21% to 62% on the CUHK03 benchmark.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/12bce6e2db10faa4f370f9e40a6084296080b5cb

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

1. Liang Zheng
2. Yi Yang
3. Qi Tian
4. Shaogang Gong
5. Tao A. Xiang
6. Xiatian Zhu
7. Wanli Ouyang
8. Xiaogang Wang

In Computer Vision and Pattern Recognition

1. Yi Yang
2. Luc Van Gool
3. Xiaogang Wang
4. Dacheng Tao
5. Ming-Hsuan Yang
6. Liang Zheng
7. Qi Tian
8. Trevor Darrell

Citations in the corpus

• Part-Aligned Bilinear Representations for Person Re-identification (2018, methodology): “The early solutions of person re-identification mainly relied on hand-crafted features [20– 23], metric learning techniques [2,24–29], and probabilistic patch matching algorithms [30– 32] to handle resolution/light/view/pose changes.”
• Domain-Adversarial Training of Neural Networks (2015, background): “Recently, several papers with significantly improved re-identification performance (Zhang and Saligrama, 2014; Zhao et al., 2014; Paisitkriangkrai et al., 2015) have been presented, with Ma et al.”
• Hierarchical Gaussian Descriptor for Person Re-identification (2016, methodology): “State-of-the-arts: In Table 2, we compare the performance of the reported results on the state-of-the-art methods, including MidLevel Filter Learning (MLFL) [47], Salience Matching (SalMatch) [45] , SCNCD [43], Semantic attribute representation [38], Metric Ensemble [31]”
• End-to-End Deep Learning for Person Search (2016, background): “…works of person re-identification focus on manually designing discriminative features [35, 11, 37], automatically learning features with convolutional neural networks (CNN) [18, 1], learning feature transforms across camera views [25, 24, 30], and learning distance metrics [40, 10, 26, 23, 21].”
• Reinforced Temporal Attention and Split-Rate Transfer for Depth-Based Person Re-identification (2017, background): “As such, there is vast literature in learning discriminative distance metrics [35,42,105,51,55,77,43,61,53,19,48,108,91,5] and discriminant subspaces [63,47,107,84,43,46,64,94,15].”
• In Defense of the Triplet Loss for Person Re-Identification (2017, methodology): “The function fθ is parametrized by θ and can be anything ranging from a linear transform [33, 19, 36, 22] to complex non-linear mappings usually represented by deep neural networks [7, 6, 8].”
• Enhancing Person Re-identification in a Self-Trained Subspace (2017, background): “Some existing works [31], [40] have investigated the effects of multiple feature representations for re-ID by a score-level fusion.”
• Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function (2016, background): “The basic idea behind metric learning is to find a mapping function from the feature space to the distance space with certain merits, such as feature vectors from the same person being closer than those from different ones.”
• Person Re-identification: Past, Present and Future (2016, background): “For example, on the most studied dataset VIPeR [82], from the year 2008 to 2016, representative works [13], [24], [30], [32], [85], [94], [95] witness a rank-1 accuracy from 12.0% in 2008 [24] to 63.9% in 2015 [94], an improvement of +51.9%.”
• Person Re-identification by Deep Learning Multi-scale Representations (2017, background): “They have achieved a sequence of continuous re-id performance boost on benchmarking datasets when integrated with various supervised matching models [22, 41, 75, 28, 68, 29, 30, 40, 72, 63, 73, 62, 80, 42, 71].”
• Joint Detection and Identification Feature Learning for Person Search (2016, methodology): “Early person re-identification methods addressed the problem by manually designing discriminative features [12,33,41], learning feature transforms across camera views [25,26,29], and learning distance metrics [11,22,24,26,45].”