Evaluating Contrastive Models for Instance-based Image Retrieval

TL;DR

This paper evaluates contrastive learning models for image retrieval, demonstrating they perform comparably or better than supervised models trained on ImageNet, despite requiring no explicit labels.

Contribution

The study provides extensive evaluation showing contrastive models are effective for image retrieval, highlighting their potential as label-free alternatives.

Findings

Contrastive models perform on-par or better than supervised baselines.

Contrastive models require no explicit supervision.

Effective for building robust image retrieval systems.

Abstract

In this work, we evaluate contrastive models for the task of image retrieval. We hypothesise that models that are learned to encode semantic similarity among instances via discriminative learning should perform well on the task of image retrieval, where relevancy is defined in terms of instances of the same object. Through our extensive evaluation, we find that representations from models trained using contrastive methods perform on-par with (and outperforms) a pre-trained supervised baseline trained on the ImageNet labels in retrieval tasks under various configurations. This is remarkable given that the contrastive models require no explicit supervision. Thus, we conclude that these models can be used to bootstrap base models to build more robust image retrieval engines.