Reinforcing Pre-trained Models Using Counterfactual Images

TL;DR

This paper introduces a framework that uses language-guided generated counterfactual images to identify and reinforce classification models, improving their robustness by fine-tuning with these counterfactuals.

Contribution

It presents a novel two-stage process for identifying model weaknesses and reinforcing models using counterfactual images generated from perturbed captions.

Findings

Fine-tuning with counterfactual images improves model robustness.

The framework effectively identifies vulnerabilities in classification models.

Experiments show significant performance gains across datasets.

Abstract

This paper proposes a novel framework to reinforce classification models using language-guided generated counterfactual images. Deep learning classification models are often trained using datasets that mirror real-world scenarios. In this training process, because learning is based solely on correlations with labels, there is a risk that models may learn spurious relationships, such as an overreliance on features not central to the subject, like background elements in images. However, due to the black-box nature of the decision-making process in deep learning models, identifying and addressing these vulnerabilities has been particularly challenging. We introduce a novel framework for reinforcing the classification models, which consists of a two-stage process. First, we identify model weaknesses by testing the model using the counterfactual image dataset, which is generated by perturbed image captions. Subsequently, we employ the counterfactual images as an augmented dataset to fine-tune and reinforce the classification model. Through extensive experiments on several classification models across various datasets, we revealed that fine-tuning with a small set of counterfactual images effectively strengthens the model.