Robust Image Ordinal Regression with Controllable Image Generation

TL;DR

This paper introduces CIG, a controllable image generation framework that enhances ordinal regression by addressing class imbalance and category overlap through generating targeted training samples, especially for minority categories.

Contribution

The novel CIG framework enables controllable image generation to improve ordinal regression performance, particularly for underrepresented categories, by separating structural and categorical information.

Findings

CIG improves performance across three ordinal regression scenarios.

The method significantly benefits minority category classification.

CIG can be integrated with existing models for enhanced results.

Abstract

Image ordinal regression has been mainly studied along the line of exploiting the order of categories. However, the issues of class imbalance and category overlap that are very common in ordinal regression were largely overlooked. As a result, the performance on minority categories is often unsatisfactory. In this paper, we propose a novel framework called CIG based on controllable image generation to directly tackle these two issues. Our main idea is to generate extra training samples with specific labels near category boundaries, and the sample generation is biased toward the less-represented categories. To achieve controllable image generation, we seek to separate structural and categorical information of images based on structural similarity, categorical similarity, and reconstruction constraints. We evaluate the effectiveness of our new CIG approach in three different image ordinal regression scenarios. The results demonstrate that CIG can be flexibly integrated with off-the-shelf image encoders or ordinal regression models to achieve improvement, and further, the improvement is more significant for minority categories.