Improving CLIP Training with Language Rewrites

TL;DR

This paper introduces LaCLIP, a method that enhances CLIP training by using language rewrites generated by large language models to diversify text descriptions, leading to significant improvements in transfer performance.

Contribution

The paper proposes a novel language augmentation technique for CLIP training using language rewrites, which improves transfer accuracy without additional computational costs.

Findings

LaCLIP outperforms CLIP in zero-shot ImageNet accuracy by up to 8.2%.

Language rewrites increase diversity of text inputs, enhancing model robustness.

The method requires no extra computation or memory during training.

Abstract

Contrastive Language-Image Pre-training (CLIP) stands as one of the most effective and scalable methods for training transferable vision models using paired image and text data. CLIP models are trained using contrastive loss, which typically relies on data augmentations to prevent overfitting and shortcuts. However, in the CLIP training paradigm, data augmentations are exclusively applied to image inputs, while language inputs remain unchanged throughout the entire training process, limiting the exposure of diverse texts to the same image. In this paper, we introduce Language augmented CLIP (LaCLIP), a simple yet highly effective approach to enhance CLIP training through language rewrites. Leveraging the in-context learning capability of large language models, we rewrite the text descriptions associated with each image. These rewritten texts exhibit diversity in sentence structure and vocabulary while preserving the original key concepts and meanings. During training, LaCLIP randomly selects either the original texts or the rewritten versions as text augmentations for each image. Extensive experiments on CC3M, CC12M, RedCaps and LAION-400M datasets show that CLIP pre-training with language rewrites significantly improves the transfer performance without computation or memory overhead during training. Specifically for ImageNet zero-shot accuracy, LaCLIP outperforms CLIP by 8.2% on CC12M and 2.4% on LAION-400M. Code is available at https://github.com/LijieFan/LaCLIP.