Restoranet: High Resolution Image Restoration with Transformers

Noise, blur, and compression artifacts commonly degrade high-resolution images, reducing visual quality and affecting various downstream applications in medical imaging, satellite imaging, and photography. However, traditional image restoration methods usually suffer from balancing accuracy and computational efficiency, and most of them cannot handle large-size high-resolution images. This paper presents an efficient transformer-based model for high-resolution image restoration, termed RestoraNet. Benefiting from the self-attention mechanism, RestoraNet effectively leverages the long-range dependencies and complex contextual information of images, leading to superior restoration performance. In this work, multiscale feature extraction, residual connections, and efficient attention modules will be incorporated into the model for enhancing detail preservation while reducing computational cost.