The Group Loss for Deep Metric Learning

Elezi, Ismail; Vascon, Sebastiano; Torcinovich, Alessandro; Pelillo, Marcello; Leal-Taixé, Laura

doi:10.1007/978-3-030-58571-6_17

Cited by 40 publications

(19 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep metric learning may be loosely stated as the task of obtaining good features for downstream tasks [35]. In this paper we wish to use deep metric learning to construct discriminative latent spaces for VAE BO.…”

Section: Deep Metric Learningmentioning

confidence: 99%

High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

Grosnit¹,

Tutunov²,

Maraval³

et al. 2021

Preprint

View full text Add to dashboard Cite

Section: Deep Metric Learningmentioning

confidence: 99%

High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

Grosnit¹,

Tutunov²,

Maraval³

et al. 2021

Preprint

View full text Add to dashboard Cite

“…Classification losses, in contrast to pairwise losses, perform the optimization independently per image. An exception is the work Elezi et al [10] where a similarity propagation module captures group interactions within the batch. Then, Cross-Entropy (CE) loss is used, which now comes with significant improvements by taking into account such interactions.…”

Section: Related Workmentioning

confidence: 99%

Recall@k Surrogate Loss with Large Batches and Similarity Mixup

Patel¹,

Tolias²,

Matas³

2021

Preprint

View full text Add to dashboard Cite

Direct optimization, by gradient descent, of an evaluation metric, is not possible when it is non-differentiable, which is the case for recall in retrieval. In this work, a differentiable surrogate loss for the recall is proposed. Using an implementation that sidesteps the hardware constraints of the GPU memory, the method trains with a very large batch size, which is essential for metrics computed on the entire retrieval database. It is assisted by an efficient mixup approach that operates on pairwise scalar similarities and virtually increases the batch size further. When used for deep metric learning, the proposed method achieves stateof-the-art results in several image retrieval benchmarks. For instance-level recognition, the method outperforms similar approaches that train using an approximation of average precision. The implementation will be made public.

show abstract

“…The success of deep neural networks in many computer vision tasks (e.g., visual recognition [12]) has led to many works on learning visual similarity using deep neural networks (i.e., deep metric learning) [2,4,5,8,9,17,19,22]. The core of these DML methods is a loss function that guides the training of networks that transform images to a compact embedding space, where similar images are mapped close together while dissimilar images are far apart.…”

Section: Related Workmentioning

confidence: 99%

“…It is often treated as a metric learning problem where the task is to represent images with compact embedding vectors such that semantically similar images are grouped together while dissimilar images are far apart in the embedding space. Inspired by the success of deep neural networks in visual recognition, convolutional neural networks have also been employed in metric learning, which is specifically called deep metric learning (DML) [2,4,5,8,17,21,22,38].…”

Section: Introductionmentioning

confidence: 99%

Hierarchical Proxy-based Loss for Deep Metric Learning

Yang¹,

Baştan²,

Zhu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Proxy-based metric learning losses are superior to pairbased losses due to their fast convergence and low training complexity. However, existing proxy-based losses focus on learning class-discriminative features while overlooking the commonalities shared across classes which are potentially useful in describing and matching samples. Moreover, they ignore the implicit hierarchy of categories in realworld datasets, where similar subordinate classes can be grouped together. In this paper, we present a framework that leverages this implicit hierarchy by imposing a hierarchical structure on the proxies and can be used with any existing proxy-based loss. This allows our model to capture both class-discriminative features and class-shared characteristics without breaking the implicit data hierarchy. We evaluate our method on five established image retrieval datasets such as In-Shop and SOP. Results demonstrate that our hierarchical proxy-based loss framework improves the performance of existing proxy-based losses, especially on large datasets which exhibit strong hierarchical structure.

show abstract

The Group Loss for Deep Metric Learning

Cited by 40 publications

References 39 publications

High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning

Recall@k Surrogate Loss with Large Batches and Similarity Mixup

Hierarchical Proxy-based Loss for Deep Metric Learning

Contact Info

Product

Resources

About