Unleashing Vision Transformer Potential In Image Quality Assessment via Global-Local Adaptive Interaction

ArXi:2605.17748v1 Announce Type: new In the field of Blind Image Quality Assessment (BIQA), accurately predicting the perceptual quality of authentically distorted images remains highly challenging due to the diverse and complex distortions present in natural environments. Although existing methods have achieved notable accuracy, their scalability is often constrained by the high cost of subjective annotation and the limited size of available datasets. Recent advances in large-scale pre-trained vision models have.