Rank-smoothed Pairwise Learning In Perceptual Quality Assessment

TL;DR

This paper introduces a rank-smoothed loss function for training deep image quality assessment models, which improves the accuracy of predicting human perceptual preferences by regularizing pairwise empirical probabilities with aggregated rankwise probabilities.

Contribution

The paper proposes a novel rank-smoothed loss that enhances deep learning models for perceptual quality assessment by incorporating global ranking information.

Findings

Rank-smoothed loss improves preference prediction accuracy.

Regularization with rankwise probabilities leads to more reliable training.

Method outperforms traditional pairwise training approaches.

Abstract

Conducting pairwise comparisons is a widely used approach in curating human perceptual preference data. Typically raters are instructed to make their choices according to a specific set of rules that address certain dimensions of image quality and aesthetics. The outcome of this process is a dataset of sampled image pairs with their associated empirical preference probabilities. Training a model on these pairwise preferences is a common deep learning approach. However, optimizing by gradient descent through mini-batch learning means that the "global" ranking of the images is not explicitly taken into account. In other words, each step of the gradient descent relies only on a limited number of pairwise comparisons. In this work, we demonstrate that regularizing the pairwise empirical probabilities with aggregated rankwise probabilities leads to a more reliable training loss. We show that training a deep image quality assessment model with our rank-smoothed loss consistently improves the accuracy of predicting human preferences.