SwinIR: Image Restoration Using Swin Transformer

SwinIR is an image restoration model that brings the Swin Transformer to a low-level vision problem long dominated by convolutional networks. The architecture has three parts: shallow feature extraction, deep feature extraction, and high-quality image reconstruction. The deep feature extraction module is built from several residual Swin Transformer blocks (RSTB), each containing multiple Swin Transformer layers plus a residual connection, letting the model capture rich features while remaining trainable.

Evaluated on three representative tasks - image super-resolution (classical, lightweight, and real-world), image denoising (grayscale and color), and JPEG compression artifact reduction - SwinIR outperforms state-of-the-art methods by up to 0.14 to 0.45 dB. It achieves this while reducing the total number of parameters by up to 67%, demonstrating that Transformer-based designs can be both more accurate and more compact for image restoration.