Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Differentiable Low-computation Global Correlation Loss for Monotonicity Evaluation in Quality Assessment

Yipeng Liu, Qi Yang, Yiling Xu

TL;DR

The paper addresses the non-differentiable nature of SROCC-based monotonicity evaluation in no-reference quality assessment by introducing a differentiable, low-cost monotonicity loss derived from a curve-based approximation of sorting. It further stabilizes global evaluation via a memory bank that stores gradient-free predictions from previous batches, aligning batch optimization with global SROCC goals. The method is evaluated on both image IQA and point-cloud QA, showing consistent improvements over strong baselines and across datasets. Overall, the work provides a practical approach to integrate global monotonicity into training for cross-domain quality assessment.

Abstract

In this paper, we propose a global monotonicity consistency training strategy for quality assessment, which includes a differentiable, low-computation monotonicity evaluation loss function and a global perception training mechanism. Specifically, unlike conventional ranking loss and linear programming approaches that indirectly implement the Spearman rank-order correlation coefficient (SROCC) function, our method directly converts SROCC into a loss function by making the sorting operation within SROCC differentiable and functional. Furthermore, to mitigate the discrepancies between batch optimization during network training and global evaluation of SROCC, we introduce a memory bank mechanism. This mechanism stores gradient-free predicted results from previous batches and uses them in the current batch's training to prevent abrupt gradient changes. We evaluate the performance of the proposed method on both images and point clouds quality assessment tasks, demonstrating performance gains in both cases.

Differentiable Low-computation Global Correlation Loss for Monotonicity Evaluation in Quality Assessment

TL;DR

The paper addresses the non-differentiable nature of SROCC-based monotonicity evaluation in no-reference quality assessment by introducing a differentiable, low-cost monotonicity loss derived from a curve-based approximation of sorting. It further stabilizes global evaluation via a memory bank that stores gradient-free predictions from previous batches, aligning batch optimization with global SROCC goals. The method is evaluated on both image IQA and point-cloud QA, showing consistent improvements over strong baselines and across datasets. Overall, the work provides a practical approach to integrate global monotonicity into training for cross-domain quality assessment.

Abstract

In this paper, we propose a global monotonicity consistency training strategy for quality assessment, which includes a differentiable, low-computation monotonicity evaluation loss function and a global perception training mechanism. Specifically, unlike conventional ranking loss and linear programming approaches that indirectly implement the Spearman rank-order correlation coefficient (SROCC) function, our method directly converts SROCC into a loss function by making the sorting operation within SROCC differentiable and functional. Furthermore, to mitigate the discrepancies between batch optimization during network training and global evaluation of SROCC, we introduce a memory bank mechanism. This mechanism stores gradient-free predicted results from previous batches and uses them in the current batch's training to prevent abrupt gradient changes. We evaluate the performance of the proposed method on both images and point clouds quality assessment tasks, demonstrating performance gains in both cases.
Paper Structure (16 sections, 12 equations, 2 figures, 4 tables)

This paper contains 16 sections, 12 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. No-Reference Image Quality Assessment (NR-IQA)
  4. No-Reference Point Cloud Quality Assessment (NR-PCQA)
  5. Loss Function Design in Model Training
  6. Preliminary
  7. Methodology
  8. Differentiable Approximation
  9. Memory Bank Mechanism
  10. Experiments
  11. Datasets
  12. Implementation Details
  13. Performance Gain
  14. Ablation Study
  15. Conclusion
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Differentiable fitting of step function. $k$ controls the steepness of the fitting function curve.
  • Figure 2: Memory mechanism of predicted quality scores for global monotonicity evaluation. In each batch, the output quality scores of the network are recorded/updated in the memory bank. When optimizing the global monotonicity loss, the current and previous batch results are input together.