Less is More: Removing Text-regions Improves CLIP Training Efficiency and Robustness

TL;DR

This paper proposes removing images with text regions during CLIP training to enhance efficiency and robustness, leading to significant accuracy improvements and protection against typographic attacks.

Contribution

It introduces a novel filtering approach to exclude text-containing images, improving CLIP training efficiency and robustness against adversarial text manipulations.

Findings

Filtering out text regions boosts classification accuracy on benchmarks.

The approach enhances robustness against typographic adversarial attacks.

Achieved top-1 accuracy of 68.78% on ImageNet-Attr, surpassing previous models.

Abstract

The CLIP (Contrastive Language-Image Pre-training) model and its variants are becoming the de facto backbone in many applications. However, training a CLIP model from hundreds of millions of image-text pairs can be prohibitively expensive. Furthermore, the conventional CLIP model doesn't differentiate between the visual semantics and meaning of text regions embedded in images. This can lead to non-robustness when the text in the embedded region doesn't match the image's visual appearance. In this paper, we discuss two effective approaches to improve the efficiency and robustness of CLIP training: (1) augmenting the training dataset while maintaining the same number of optimization steps, and (2) filtering out samples that contain text regions in the image. By doing so, we significantly improve the classification and retrieval accuracy on public benchmarks like ImageNet and CoCo. Filtering out images with text regions also protects the model from typographic attacks. To verify this, we build a new dataset named ImageNet with Adversarial Text Regions (ImageNet-Attr). Our filter-based CLIP model demonstrates a top-1 accuracy of 68.78\%, outperforming previous models whose accuracy was all below 50\%.