.png){kind=icon}
.png){kind=icon}
Return
SRTransGAN: Image Super-Resolution Using Transformer Based Generative Adversarial Network
.png){kind=icon}
PRE
AI
DOI：10.1007/978-3-031-93694-4_22.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Image super-resolution is a critical task that aims to synthesize high-resolution images from low-resolution images, addressing limitations in applications such as low-resolution object recognition and medical image enhancement. Generative adversarial network (GAN) based methods have been the state-of-the-art in image super-resolution, utilising convolutional neural networks (CNNs) for both generator and discriminator networks. However, CNNs are not optimal for exploiting global information, unlike transformers, a recent breakthrough in deep learning due to their self-attention mechanism. Inspired by the success of transformers in language and vision applications, we propose a novel image super-resolution method called SRTransGAN, which leverages a transformer-based GAN architecture. Our approach employs a transformer-based encoder-decoder network as a generator to produce 2x and 4x high-resolution images. Our proposed SRTransGAN outperforms existing methods by 4.38% on average, as measured by PSNR and SSIM scores. We also demonstrate the effectiveness of our method through saliency map analysis. Our method provides a promising new approach for image super-resolution, highlighting the potential of transformerbased architectures in addressing this critical problem. The code used in the paper is publicly available at https://github.com/nbaghel777/SRTransGAN.
Keywords:
High Resolution
Transformer
GAN
Efficient Network
Encoder-Decoder
Super-Resolution
Image Quality
Upscaling Techniques
Journal
C
IF:
0
Papers:
32
Citations:
0