SwinIR: Image Restoration Using Swin Transformer
SwinIR: восстановление изображений с использованием Swin Transformer
2021-10-01
SCID: 54.1/b8c5kpmy
Discuss with AI
JPEG compression artifact reductionSwin TransformerSwinIRimage denoisingimage super-resolutionresidual Swin Transformer block
Figures from the paper
Abstract (AI)
Image restoration is a long-standing low-level vision problem that aims to restore high-quality images from low-quality images (e.g., downscaled, noisy and compressed images). While state-of-the-art image restoration methods are based on convolutional neural networks, few attempts have been made with Transformers which show impressive performance on high-level vision tasks. In this paper, we propose a strong baseline model SwinIR for image restoration based on the Swin Transformer. SwinIR consists of three parts: shallow feature extraction, deep feature extraction and high-quality image reconstruction. In particular, the deep feature extraction module is composed of several residual Swin Transformer blocks (RSTB), each of which has several Swin Transformer layers together with a residual connection. We conduct experiments on three representative tasks: image super-resolution (including classical, lightweight and real-world image super-resolution), image denoising (including grayscale and color image denoising) and JPEG compression artifact reduction. Experimental results demonstrate that SwinIR outperforms state-of-the-art methods on different tasks by up to 0.14∼0.45dB, while the total number of parameters can be reduced by up to 67%.
Key Findings
1
Each RSTB contains several Swin Transformer layers with residual connections for deep feature extraction.
2
SwinIR architecture comprises shallow feature extraction, deep feature extraction using residual Swin Transformer blocks (RSTB), and high-quality image reconstruction.
3
SwinIR can reduce total parameter count by up to 67% compared to prior state-of-the-art methods.
4
SwinIR is a strong baseline image restoration model built on the Swin Transformer architecture.
5
SwinIR outperforms state-of-the-art methods across tasks by up to 0.14–0.45 dB.
6
SwinIR was evaluated on image super-resolution (classical, lightweight, real-world), image denoising (grayscale and color), and JPEG compression artifact reduction.
Research Object
SwinIR image restoration model based on the Swin Transformer applied to low-quality images
Research Subject
Restoration performance (super-resolution, denoising, JPEG artifact reduction), model architecture (shallow/deep feature extraction with residual Swin Transformer blocks), and parameter-efficiency compared to state-of-the-art methods
Publication Details
Publication Date
2021-10-01
Journal
Publisher
ISSN
Cited by
5078
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai9
Exploiting Generative AI to Scale up Intelligent Tutoring Systems2023
Swin Transformer: Hierarchical Vision Transformer using Shifted Windows2021
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale2020
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network2017
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising2017
Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers2021
Deep Learning for Generic Object Detection: A Survey2019
Uformer: A General U-Shaped Transformer for Image Restoration2022
Pre-Trained Image Processing Transformer2021