Large scale metric learning from equivalence constraints

By Martin Köstinger, Martin Hirzer, Paul Wohlhart, Peter M. Roth, Horst Bischof. 2012

1,697 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Unsupervised Feature Learning via Non-parametric Instance Discrimination (2018, background): “There have been extensive studies on metric learning [15, 33].”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, background): “Traditional metric learning [36, 33, 12, 38] usually learns a matrix A for a distance metric ‖x1−x2‖A= √ (x1−x2)A(x1−x2) upon the given features x1,x2.”
Semantic Instance Segmentation with a Discriminative Loss Function (2017, background): “ handle occlusions and produce more precise segmentations. Loss function Our loss function is inspired by earlier works on distance metric learning, discriminative loss functions and siamese networks [8,10,16,39,21]. Most similar to our loss function, Weinberger et al. [39] propose to learn a distan”
Distance-Based Image Classification: Generalizing to New Classes at Near-Zero Cost (2013, neutral): “[21], [22], [23].”
Multi-scale Deep Learning Architectures for Person Re-identification (2017, background): “earance representation and employ distance metric learning models as matching functions. They focus on either designing cross-view robust features [7,13,26,62,32], or learning robust distance metrics [31,33,63,25,49, 60,38,32], or both [24,32,57,59]. Recently, inspired by the success of convolutiona”
Person Re-Identification by Camera Correlation Aware Feature Augmentation (2017, methodology): “[13], PCCA [6], LFDA [10], XQDA [12], PSD[15], Metric”
Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification (2016, methodology): “curacies of the proposed model (marked as “Ours (Pairwise/Triplet/Combined, Pretrain)”) and the other state−of− the−art person re−identification methods, including FPNN [19], Euclidean distance, ITML [5], LMNN [35], RANK [27], LDML [13], SDALF [8], eSDC [41], KISSME [16], and the work by Ahmed et al”
Deeply-Learned Part-Aligned Representations for Person Re-identification (2017, neutral): “ on CUHK03 for manually labeled human boxes. Method rank-1 rank-5 rank-10 rank-20 BoW [64] (ICCV15) 18:9 36:2 46:8 - F LOMO [26] (CVPR15) 52:2 82:2 92:1 96:3 GOG [31] (CVPR16) 67:3 91:0 96:0 - KISSME [20] (CVPR12) 47:9 69:3 78:9 87:0 SSSVM [58] (CVPR16) 57:0 84:8 92:5 96:4 M DNS [57] (CVPR16) 58:9 8”
Enhancing Person Re-identification in a Self-Trained Subspace (2017, methodology): “During the past decades, many person re-ID algorithms [2], [3], [11], [13], [14], [21]–[26] have been proposed.”
Clothing Attributes Assisted Person Reidentification (2015, methodology): “9, the performance of KISSME is better than conventional SVM, but not as good as the proposed c-LSVM.”
Cross Dataset Person Re-identification (2014, methodology): “Koestinger et al.[10] proposed the KISSME method to learn a distance metric from equivalence constraints based on a statistical inference perspective.”
Quality Aware Network for Set to Set Recognition (2017, neutral): “.3 79.4 88.2 91.9 CNN+RNN [36] 58 84 91 96 STFV3D [22] 37.0 64.3 77.0 86.9 TDL [40] 56.3 87.6 95.6 98.3 eSDC [34] 41.3 63.5 72.7 83.1 DVR [34] 39.5 61.1 71.7 81.0 LFDA [25] 32.9 68.5 82.2 92.6 KISSME [16] 36.5 67.8 78.8 87.1 LADF [21] 39.0 76.8 89.0 96.8 TopRank [19] 22.5 56.1 72.7 85.9 Table 2. Com”
Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function (2016, methodology): “2 KISSME[21] 19.”
Learning Deep Context-Aware Features over Body and Latent Parts for Person Re-identification (2017, methodology): “[7, 10, 18, 19, 22, 27, 41–43, 50, 51], or learning an effective distance metric [2, 16, 21, 22, 29, 47, 57].”
Fine-grained visual categorization via multi-stage metric learning (2014, positive): “metric that keeps data points of the same class close to each other and data points from different classes far apart. It has been successfully applied to several important problems in computer vision [4,5,10,12] and the new space spanned by the learned metric serves much better than the original spa”
Frontal to profile face verification in the wild (2016, background): “Researchers have also developed other metric learning approaches [13], [18] along with Deep metric learning approaches [15] [16].”
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 S”
Learning Local Image Descriptors with Deep Siamese and Triplet Convolutional Networks by Minimizing Global Loss Functions (2015, methodology): “The method above has been extended in many ways, such that: 1) it can handle multimodal distributions in [31]; 2) it optimises K nearest neighbour classification, which is formulated as a softmax loss minimisation and estimates a linear transform with eigenvalue decomposition [11]; 3) it optimises a”
Scalable Person Re-identification: A Benchmark (2015, methodology): “We then apply various metric learning methods [34, 6, 17] on BoW (after PCA to 100dim).”