A Survey on Ordinal Regression: Applications, Advances and Prospects

TL;DR

The paper tackles ordinal regression by leveraging the inherent order in category labels across diverse domains. It introduces a three-paradigm taxonomy—Continuous Space Discretization, Distribution Ordering Learning, and Ambiguous Instance Delving—to organize how methods transform, constrain, and refine ordinal predictions. The survey covers image-level and pixel-level tasks, probabilistic and label-distribution approaches, and CLIP-based pre-training, mapping techniques to concrete datasets and applications. It also discusses challenges such as ambiguous samples, inconsistent label distributions, and out-of-domain generalization, and outlines promising directions including vision-language models and ordinal foundation models for improved zero-shot and few-shot generalization.

Abstract

Ordinal regression refers to classifying object instances into ordinal categories. Ordinal regression is crucial for applications in various areas like facial age estimation, image aesthetics assessment, and even cancer staging, due to its capability to utilize ordered information effectively. More importantly, it also enhances model interpretation by considering category order, aiding the understanding of data trends and causal relationships. Despite significant recent progress, challenges remain, and further investigation of ordinal regression techniques and applications is essential to guide future research. In this survey, we present a comprehensive examination of advances and applications of ordinal regression. By introducing a systematic taxonomy, we meticulously classify the pertinent techniques and applications into three well-defined categories based on different strategies and objectives: Continuous Space Discretization, Distribution Ordering Learning, and Ambiguous Instance Delving. This categorization enables a structured exploration of diverse insights in ordinal regression problems, providing a framework for a more comprehensive understanding and evaluation of this field and its related applications. To our best knowledge, this is the first systematic survey of ordinal regression, which lays a foundation for future research in this fundamental and generic domain.

Figures (2)

• Figure 1: Demonstrating the applications, insights, and significances of ordinal regression. Many real-world phenomena naturally provide ordinal data such as the age of a person, the level of a disease, and the depth of the radar. For the prediction of ordered data, ordinal regression was explored recently mainly in three aspects: Continuous Space Discretization, Distribution Ordering Learning, and Ambiguous Instance Delving. Importantly, the study of ordinal regression has three main significances. (1) Ordinal regression methods can be applied to broader applications since the underlying ordered logic is similar. (2) Ordinal regression methods can fully utilize meaningful order information in ordinal data and even there is a potential to train a universal model on all ordered data. (3) Ordinal regression methods can be more interpretable since they consider order relationships between categories which may contain causal information and trends behind the data.
• Figure 2: A taxonomy of ordinal regression with representative examples.