Metric Learning for Large Scale Image Classification: Generalizing to New Classes at Near-Zero Cost

By Thomas Mensink, Jakob Verbeek, Florent Perronnin, Gabriella Csurka. 2012

339 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

ImageNet Large Scale Visual Recognition Challenge (2014, background): “…Tatsuya Harada†‡, Yasuo Kuniyoshi† (Harada and Kuniyoshi, 2012) LEAR 34.5 - LEAR INRIA Grenoble†, TVPA Xerox Research Centre Europe‡ Thomas Mensink†‡, Jakob Verbeek†, Florent Perronnin‡, Gabriela Csurka‡ (Mensink et al., 2012) VGG 27.0 50.0 University of Oxford Karen Simonyan, Yusuf Aytar,…”
Learning Deep Structure-Preserving Image-Text Embeddings (2015, background): “Such within-view structure preservation constraints have been extensively explored in the metric learning literature [19, 32, 40, 41, 48, 53].”
Towards Open Set Deep Networks (2015, methodology): “To recognize outliers using AVs, we adapt the concepts of Nearest Class Mean [29, 12] or Nearest Non-Outlier [1] and apply them per class within the activation vector, as a first approximation.”
iCaRL: Incremental Classifier and Representation Learning (2016, background): “Despite (or because of) its simplicity, NCM has been shown to work well and be more robust than standard parametric classifiers in an incremental learning setting [23, 24, 31].”
Dynamic Few-Shot Visual Learning Without Forgetting (2018, methodology): “The proposed modification in the architecture of a classification ConvNet allows to unify the recognition of base and novel categories without significantly altering the classification pipeline for the recognition of base categories (in contrast to [11, 17]).”
Learning to Compare: Relation Network for Few-Shot Learning (2017, background): “Also related are conventional metric learning approaches [26, 7] that focus on learning a shallow (linear) Mahalanobis metric for a fixed feature representa-”
Three scenarios for continual learning (2019, methodology): “This method uses a neural network for feature extraction and performs classification based on a nearest-class-mean rule [23] in that feature space, whereby the class means are calculated from the stored data.”
Deep Metric Learning via Lifted Structured Feature Embedding (2015, background): “Figure 2c illustrates the concept of of transforming a training batch of examples to a fully connected dense matrix of pairwise distances.”
The limits and potentials of deep learning for robotics (2018, background): “Current techniques for class-incremental learning [75], [90] still rely on supervision in the sense that the user has to speciﬁcally tell the system which samples are new data and therefore should be incorporated.”
Towards Open World Recognition (2014, positive): “Recently the technique has been adapted for use in larger scale vision problems [48, 47, 31, 36], with the most recent and most accurate approaches combining NCM with metric learning [31] and with random forests[36].”
Revisiting Distillation and Incremental Classifier Learning (2018, neutral): “[10] demonstrated that by fixing the representation”
Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (2017, background): “[55] does not use side information however it requires one-shot”
M2KD: Multi-model and Multi-level Knowledge Distillation for Incremental Learning (2019, methodology): “An external nearest class mean classifier [23] is applied at inference.”
A Strategy for an Uncompromising Incremental Learner (2017, methodology): “The NCM approach has also been successfully applied to random forest based models for incremental learning in [25].”
Zero-Shot Learning by Convex Combination of Semantic Embeddings (2013, methodology): “Recently, zero-shot learning [10, 14] has received a growing amount of attention [16, 11, 6, 18].”
Learning Two-Branch Neural Networks for Image-Text Matching Tasks (2017, background): “Such within-view neighborhood preservation constraints have been extensively explored in the metric learning literature [38], [39], [40], [41], [42], [43].”
Label-Embedding for Image Classification (2015, neutral): “Possible sources of prior information include attributes [30], [18], [43], [47], [46], semantic class taxonomies [46], [39], class-to-class similarities [47], [70], text features [43], [47], [46], [57], [20] or class co-occurrence statistics [37].”