Deep Metric Learning via Facility Location

By Hyun Oh Song, Stefanie Sabrina Jegelka, Vivek Rathod, Kevin P. Murphy. 2016

304 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

Citations in the corpus

Unsupervised Embedding Learning via Invariant and Spreading Instance Feature (2019, neutral): “5 Clustering[38] 48.”
Smart Mining for Deep Metric Learning (2017, methodology): “∗The first two authors contributed equally to this work the presence of an extremely large number of classes (more than 10(5) classes) and low number of samples per class (in [10(1), 10(2)]), where the implementation of traditional classifiers becomes challenging [19, 13].”
Significance of Softmax-Based Features in Comparison to Distance Metric Learning-Based Features (2017, methodology): “End-to-end distance metric learning is a typical approach to constructing a feature extractor using convolutional neural networks and has been the focus of numerous studies [7], [8], [9], [10], [11].”
A Survey on Multi-View Clustering (2017, unknown): “The works in [18], [156], [157] borrowed the supervised deep learning idea to perform supervised clustering.”
Learning an Evolutionary Embedding via Massive Knowledge Distillation (2020, background): “Beyond these local sampling methods deﬁned on one batch, Song et al. (2017) propose a new metric learning scheme which is aware of the global structure of the embedding space.”
Learning to Discover Novel Visual Categories via Deep Transfer Clustering (2019, background): “Our work is also related to metric learning [29, 30, 31] and domain adaptation [36].”
Representation Learning for Visual-Relational Knowledge Graphs (2017, background): “Recent work has extended this idea to directly optimize a clustering quality metric [43].”
SoftTriple Loss: Deep Metric Learning Without Triplet Sampling (2019, unknown): “For SemiHard [21], we report the result recorded in [23].”
Memory-Based Neighbourhood Embedding for Visual Recognition (2019, background): “Images from the same or related classes are desired to be mapped to nearby points on a manifold, which is critical to many applications like few-shot learning [41, 31, 37], visual search [7, 13, 45], face/person recognition [20, 26, 43, 2] and fine-grained retrieval [19, 18].”
Answering Visual-Relational Queries in Web-Extracted Knowledge Graphs (2017, methodology): “Recent work has extended this idea to directly optimize a clustering quality metric [Song et al., 2017].”
Classification is a Strong Baseline for Deep Metric Learning (2018, result): “Interestingly, this contra-dicts to reported behaviors for previous non-parametric metric learning methods [9, 18, 19], showing that dimensionality does not signiﬁcantly affect retrieval performance.”
Heated-Up Softmax Embedding (2018, methodology): “For fair comparison, only the methods using GoogleNetV1 as base network and Euclidean distance as the final evaluation metric are listed: [1] Triplet learning with semi-hard negative mining (Schroff et al., 2015), [2] Lifted structured loss (Song et al., 2016), [3] Learnable Structured Clustering (S”
Deep Metric Learning: A Survey (2019, background): “Part loss value was calculated for each part of an image divided into five parts to evaluate each of the local patterns as a separate part Loss functions are one of the most important parts of the deep metric learning models.”
Identifying Ambiguous Similarity Conditions via Semantic Matching (2022, background): “Types of loss functions are proposed [2,9,10,22,25,30,31] to take full advantage of the similarity comparisons between instances in a mini-batch, e.”
Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning (2019, unknown): “1 10 100 1000 Clustering 64 [30] 67.0 83.7 93.2 - ProxyNCA 64 [24] 73.”
Circle Loss: A Unified Perspective of Pair Similarity Optimization (2020, methodology): “On these datasets, majority methods [19, 18, 3, 20, 13, 34] adopt the encouraged setting of learning with pair-wise labels.”