Pedestrian Recognition with a Learned Metric

By Mert Dikmen, Emre Akbas, Thomas S. Huang, Narendra Ahuja. Asian Conference on Computer Vision, 2010

353 citations recorded by the source corpus.

This paper presents a new method for viewpoint invariant pedestrian recognition problem. We use a metric learning framework to obtain a robust metric for large margin nearest neighbor classification with rejection (i.e., classifier will return no matches if all neighbors are beyond a certain distance). The rejection condition necessitates the use of a uniform threshold for a maximum allowed distance for deeming a pair of images a match. In order to handle the rejection case, we propose a novel cost similar to the Large Margin Nearest Neighbor (LMNN) method and call our approach Large Margin Nearest Neighbor with Rejection (LMNN-R). Our method is able to achieve significant improvement over previously reported results on the standard Viewpoint Invariant Pedestrian Recognition (VIPeR [1]) dataset.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/0b62ece314846fa257d76e84dd9d002d1fcd21ae

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

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Attributes Driven Multi-Camera Person Re-identification (2016, background): “…focus on two aspects: a) extracting and coding local invariant features to represent the visual appearance of a person [1,2,3,4,5,6] and b) learning a discriminative distance metric hence the distance of features from the same person can be smaller [7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23].”
PCCA: A new approach for distance learning from sparse pairwise constraints (2012, methodology): “Large Margin Component Analysis (LMCA) [27], invariant LMNN [19] and LMNN-R [8].”
Saliency Weighted Features for Person Re-identification (2014, methodology): “At this rank our method outperforms LMNN-R [11] and meets the performance of KISSME [25] which have a recognition rate of 68.”
Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification (2016, background): “Dikmen et al . improved LMNN by adding the option of rejection [6].”
Local Fisher Discriminant Analysis for Pedestrian Re-identification (2013, methodology): “The VIPER experiment shows that at rank 1, LF was able to achieve 1.68% improvement over eLDFV, 4.37% improvement over KISSME and 5.9% over LMNN-R.”
Pose-Driven Deep Convolutional Model for Person Re-identification (2017, background): “them [11, 6, 33, 29, 60, 51, 64]; (b) closing the gap between a person’s different features by learning a discriminative distance metric [32, 9, 17, 36, 55, 23, 54, 30, 26, 65, 50, 3, 27, 4, 39, 28, 10, 37, 59].”
Cross Dataset Person Re-identification (2014, methodology): “Subsequently, in [7], a method similar to LMNN called Large Margin Nearest Neighbor with Rejection (LMNN-R) was proposed and achieved significant improvement.”
Unsupervised Cross-Dataset Transfer Learning for Person Re-identification (2016, methodology): “To address these challenges, most existing research efforts on Re-ID [25, 15, 16, 4, 20, 37, 39, 36, 23, 31, 2] are based on supervised learning.”
Unsupervised Salience Learning for Person Re-identification (2013, background): “Most existing works [25, 15, 8, 29, 16, 24] handle the problem of cross-view variations and extract discriminative features by employing supervised models, which require training data with identity labels.”
Person Re-identification by Salience Matching (2013, methodology): “PRDC [23], LMNNR [4], PCCA [18], and attribute-based PRDC (aPRDC) [13], a boosting approach (ELF) [8] and Rank SVM (RankSVM) [19].”
Scalable Person Re-identification: A Benchmark (2015, background): “In discriminative models [28, 12, 7, 20, 3], classic SVM (or the RankSVM [28, 40]) and boosting [11, 30] are popular choices.”
Learning Locally-Adaptive Decision Functions for Person Verification (2013, background): “Recently, metric learning (ML) approaches [26] have been applied to person veriﬁcation [12, 19, 6, 2].”
Local Descriptors Encoded by Fisher Vectors for Person Re-identification (2012, background): “PCCA [20] reports state-of-the-art results for person re-identiﬁcation, improving over Maximally Collapsing Classes [27], ITML [18] or LMNN-R [28].”
Large scale metric learning from equivalence constraints (2012, background): “It can significantly improve results for tracking [22], image retrieval [11], face identification [9], clustering [21], or person re-identification [4].”
Relaxed Pairwise Learned Metric for Person Re-identification (2012, methodology): “Metric learning for person re-identification has been previously addressed [10, 15].”
Multi-Task Learning with Low Rank Attribute Embedding for Person Re-Identification (2015, methodology): “We compare our algorithm with 11 learning-based methods (2): Relaxed Pairwise Metric Learning (RPML) [17], Probabilistic Relative Distance Comparison (PRDC) [50], RankSVM (RSVM) [38], Salmatch [47], LMF [48], Pairwise Constrained Component Analysis (PCCA) [11], regularized PCCA (rPCCA) [42], Keep It”
Multi-Task Learning with Low Rank Attribute Embedding for Multi-Camera Person Re-Identification (2018, methodology): “We compare our algorithm with 11 supervised learning-based methods: Relaxed Pairwise Metric Learning (RPML) [8], Probabilistic Relative Distance Comparison (PRDC) [15], RankSVM (RSVM) [71], Salmatch [69], LMF [30], Pairwise Constrained Component Analysis (PCCA) [7], regularized PCCA (rPCCA) [12], Ke”
DeepReID: Deep Filter Pairing Neural Network for Person Re-identification (2014, background): “The purpose of person re-identification is to match pedestrians observed in non-overlapping camera views with visual features [13, 9, 35, 1, 7, 3, 51, 20, 19, 16, 29, 26, 24, 48, 2, 41].”
Learning Mid-level Filters for Person Re-identification (2014, background): “Existing person re-identiﬁcation approaches work on three aspects: distance learning [29, 4, 12, 18, 30, 16, 14], feature design and selection [5, 3, 17, 15, 28, 7, 20, 25], and mid-level feature learning [11, 22, 10, 13].”