Significance of Softmax-based Features in Comparison to Distance Metric Learning-based Features

TL;DR

This paper compares softmax-based features and distance metric learning features in deep networks, showing that softmax features can be more effective when trained on large datasets, challenging the assumption that DML always outperforms softmax.

Contribution

The study provides an equitable comparison between softmax-based and DML-based features, highlighting the competitive performance of softmax features on large datasets.

Findings

Softmax features perform as well as or better than DML features on large datasets.

Proper evaluation of deep features should include softmax-based features.

Softmax features are often overlooked in performance assessments.

Abstract

The extraction of useful deep features is important for many computer vision tasks. Deep features extracted from classification networks have proved to perform well in those tasks. To obtain features of greater usefulness, end-to-end distance metric learning (DML) has been applied to train the feature extractor directly. However, in these DML studies, there were no equitable comparisons between features extracted from a DML-based network and those from a softmax-based network. In this paper, by presenting objective comparisons between these two approaches under the same network architecture, we show that the softmax-based features perform competitive, or even better, to the state-of-the-art DML features when the size of the dataset, that is, the number of training samples per class, is large. The results suggest that softmax-based features should be properly taken into account when evaluating the performance of deep features.