GMC-IQA: Exploiting Global-correlation and Mean-opinion Consistency for No-reference Image Quality Assessment

TL;DR

The paper addresses the misalignment between training objectives and GCC-based evaluation metrics in no-reference image quality assessment. It introduces a Global Correlation Consistency (GCC) loss built on a differentiable pairwise preference-based rank estimation and a queue mechanism to approximate full-dataset correlations, coupled with a Mean-opinion Network (MoNet) that aggregates diverse human opinions via multi-view attention. The GMC-IQA framework demonstrates state-of-the-art performance on multiple authentic IQA datasets and shows strong cross-dataset generalization, while the GCC loss proves adaptable to various architectures and training regimes. This approach enhances training stability and convergence, enabling more robust and reliable NR-IQA in real-world distortions.

Abstract

Due to the subjective nature of image quality assessment (IQA), assessing which image has better quality among a sequence of images is more reliable than assigning an absolute mean opinion score for an image. Thus, IQA models are evaluated by global correlation consistency (GCC) metrics like PLCC and SROCC, rather than mean opinion consistency (MOC) metrics like MAE and MSE. However, most existing methods adopt MOC metrics to define their loss functions, due to the infeasible computation of GCC metrics during training. In this work, we construct a novel loss function and network to exploit Global-correlation and Mean-opinion Consistency, forming a GMC-IQA framework. Specifically, we propose a novel GCC loss by defining a pairwise preference-based rank estimation to solve the non-differentiable problem of SROCC and introducing a queue mechanism to reserve previous data to approximate the global results of the whole data. Moreover, we propose a mean-opinion network, which integrates diverse opinion features to alleviate the randomness of weight learning and enhance the model robustness. Experiments indicate that our method outperforms SOTA methods on multiple authentic datasets with higher accuracy and generalization. We also adapt the proposed loss to various networks, which brings better performance and more stable training.

Figures (7)

• Figure 1: Network architecture of the proposed mean-opinion network (left) and multi-view attention learning module (right).
• Figure 2: Comparison between the MSE loss and the GMC loss under different learning rates.
• Figure 3: SROCC curves of the MoNet trained with the MSE and the GMC loss on the validation set of Koniq and SPAQ datasets.
• Figure 4: The impact of the MAL's number on the performance of our model on three datasets.
• Figure 5: The gMAD competition against HyperIQA and MANIQA on the Koniq dataset.