Performance of SOTA Underwater Image Enhancement Models on Underwater Images with High Turbidity

In this paper, we analyse the performance of the state of the art image enhancement models U- Transformer, FUnIE-GAN, MuLA-GAN, and diffusion based models on a high-turbidity underwater images. Pre-trained models were used where available, or otherwise trained using original repository parameters. Each model's robustness under challenging under-water conditions were evaluated. The performance of the models were analysed using no reference metrics such as UIQM, UCIQE and NIQE. MuLA-GAN performed well with UIQM = 2.6049, UCIQE = 0.5989, and NIQE = 4.0721, while U-Transformer scored the highest UIQM at 2.7603. However, none consistently excelled across all metrics highlighting current models' limitations in high-turbidity settings. Diffusion-based models, particularly Diffusion SR, also showed competitive super-resolution performance. The study highlights the need for further model development and generalizability to address specific challenges in underwater image enhancement.