Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data

This paper introduces Real-ESRGAN, an AI model designed to enhance the resolution of real-world images and videos. It addresses the challenge of training super-resolution models without high-quality real data by using purely synthetic data during training. This allows the model to better handle complex and unknown degradations often found in real images, leading to more effective upscaling and quality improvement.