UNIQAT : An open‐source toolkit for reproducible image quality assessment in marine surveys

Abstract Modern marine surveys generate tens of thousands of underwater images per expedition, yet many are unusable due to image quality deterioration. Manual quality screening is labour‐intensive, subjective and inconsistent across observers and teams, hindering reproducible science, creating large backlogs of unprocessed imagery, and limiting real‐time algorithmic and robotic applications. We present UNIQAT (UNderwater Image Quality Assessment Toolkit), an open‐source Python toolkit for automated underwater image quality assessment. The toolkit computes reference‐free quality metrics across nine categories of image characteristics relevant to underwater images, such as colour cast, turbidity and contrast loss. Metrics can be combined with user‐defined weights into task‐specific quality scores tailored to different ecological applications. High‐performance computing (HPC)‐aware multiprocessing and optional deep learning architectures for metric approximation support processing at scales matching modern survey data collection. We evaluated UNIQAT on 130,845 images from nine coral reef surveys conducted by the Australian Institute of Marine Science; every result reported in the main text was produced by the traditional metrics pipeline, with no deep learning model involved, and a separate training demonstration is reported in the . The traditional metrics pipeline, run as 48 CPU workers on a 64‐core cluster node, processed 7.9 4K‐images/s, with processing time scaling linearly with image count ( r 2 = 0.98). The resulting quality scores captured expected patterns of image modulation based on the environmental context of the imaged scenes: All nine surveys showed negative depth–quality correlations ( r = −0.16 to −0.89), consistent with the depth‐dependent deterioration of image quality in underwater environments. Cross‐survey metric profiles were highly consistent (mean Spearman ρ = 0.99 across all 36 pairwise comparisons, minimum 0.96), confirming measurement stability across different dates, locations, camera systems and environmental conditions. Synthesis. UNIQAT can help deliver objective and tunable image quality metrics in a versatile toolkit that enables transparent reporting and supports unsupervised quality assessment for applications in marine monitoring and robotics.