Deep Relational Metric Learning
This addresses image clustering and retrieval by better modeling intraclass variations, though it appears incremental as an enhancement to existing deep metric learning approaches.
The paper tackles the problem that conventional metric learning losses suppress intraclass variations that could help identify unseen classes, proposing a deep relational metric learning framework that adaptively learns an ensemble of features and uses relational inference to integrate them. Experiments on CUB-200-2011, Cars196, and Stanford Online Products datasets show the framework improves existing methods and achieves competitive results.
This paper presents a deep relational metric learning (DRML) framework for image clustering and retrieval. Most existing deep metric learning methods learn an embedding space with a general objective of increasing interclass distances and decreasing intraclass distances. However, the conventional losses of metric learning usually suppress intraclass variations which might be helpful to identify samples of unseen classes. To address this problem, we propose to adaptively learn an ensemble of features that characterizes an image from different aspects to model both interclass and intraclass distributions. We further employ a relational module to capture the correlations among each feature in the ensemble and construct a graph to represent an image. We then perform relational inference on the graph to integrate the ensemble and obtain a relation-aware embedding to measure the similarities. Extensive experiments on the widely-used CUB-200-2011, Cars196, and Stanford Online Products datasets demonstrate that our framework improves existing deep metric learning methods and achieves very competitive results.