Underwater Image Super-Resolution using Generative Adversarial Network-based Model

Alireza Aghelan

Submitted on 7 November 2022, last revised on 27 November 2022


Single image super-resolution (SISR) methods can enhance the resolution and quality of underwater images. Enhancing the resolution of underwater images leads to better performance of autonomous underwater vehicles (AUVs). In this work, we fine-tune the Real-Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) model to increase the resolution of underwater images. In our proposed approach, the pre-trained generator and discriminator networks of the Real-ESRGAN model are fine-tuned using underwater image datasets. We used USR-248 and UFO-120 datasets to fine-tune the Real-ESRGAN model. Our fine-tuned model produces images with better resolution and quality compared to the original model.


Subjects: Computer Science - Computer Vision and Pattern Recognition; Electrical Engineering and Systems Science - Image and Video Processing