Unified Contrastive Learning in Image-Text-Label Space

TL;DR

This paper introduces Unified Contrastive Learning (UniCL), a novel approach that combines supervised image-label data and language-image contrastive learning into a single framework, enhancing zero-shot and supervised recognition performance.

Contribution

The work proposes a new unified formulation and learning paradigm that effectively integrates different data sources into a common space for improved image recognition.

Findings

Achieves up to 9.2% and 14.5% gains in zero-shot recognition benchmarks.

Boosts linear probe performance by 7.3% and 3.4%.

Rivals supervised learning methods on image-label datasets.

Abstract

Visual recognition is recently learned via either supervised learning on human-annotated image-label data or language-image contrastive learning with webly-crawled image-text pairs. While supervised learning may result in a more discriminative representation, language-image pretraining shows unprecedented zero-shot recognition capability, largely due to the different properties of data sources and learning objectives. In this work, we introduce a new formulation by combining the two data sources into a common image-text-label space. In this space, we propose a new learning paradigm, called Unified Contrastive Learning (UniCL) with a single learning objective to seamlessly prompt the synergy of two data types. Extensive experiments show that our UniCL is an effective way of learning semantically rich yet discriminative representations, universally for image recognition in zero-shot, linear-probe, fully finetuning and transfer learning scenarios. Particularly, it attains gains up to 9.2% and 14.5% in average on zero-shot recognition benchmarks over the language-image contrastive learning and supervised learning methods, respectively. In linear probe setting, it also boosts the performance over the two methods by 7.3% and 3.4%, respectively. Our study also indicates that UniCL stand-alone is a good learner on pure image-label data, rivaling the supervised learning methods across three image classification datasets and two types of vision backbones, ResNet and Swin Transformer. Code is available at https://github.com/microsoft/UniCL.