Instance Cross Entropy for Deep Metric Learning

TL;DR

This paper introduces Instance Cross Entropy (ICE), a novel loss function for deep metric learning that leverages structured semantic similarity, is scalable, and allows seamless sample reweighting, showing superior performance on benchmarks.

Contribution

The paper proposes ICE, a new probabilistic loss function for deep metric learning that improves scalability and reweighting capabilities over existing methods.

Findings

ICE outperforms existing loss functions on three benchmarks.

ICE is scalable to large datasets and independent of training set size.

Seamless sample reweighting enhances learning effectiveness.

Abstract

Loss functions play a crucial role in deep metric learning thus a variety of them have been proposed. Some supervise the learning process by pairwise or tripletwise similarity constraints while others take advantage of structured similarity information among multiple data points. In this work, we approach deep metric learning from a novel perspective. We propose instance cross entropy (ICE) which measures the difference between an estimated instance-level matching distribution and its ground-truth one. ICE has three main appealing properties. Firstly, similar to categorical cross entropy (CCE), ICE has clear probabilistic interpretation and exploits structured semantic similarity information for learning supervision. Secondly, ICE is scalable to infinite training data as it learns on mini-batches iteratively and is independent of the training set size. Thirdly, motivated by our relative weight analysis, seamless sample reweighting is incorporated. It rescales samples' gradients to control the differentiation degree over training examples instead of truncating them by sample mining. In addition to its simplicity and intuitiveness, extensive experiments on three real-world benchmarks demonstrate the superiority of ICE.