{
 "id": "epfaa4ky63",
 "slug": "super-resolution-generative-adversarial-network",
 "title": "Super-resolution generative adversarial network",
 "updated": "2026-10-15",
 "topic_path": [
  {
   "id": "technology",
   "label": "Technology and the built world",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology"
  },
  {
   "id": "technology.computing",
   "label": "Computing and digital systems",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing"
  },
  {
   "id": "technology.computing.ai",
   "label": "Artificial intelligence and data",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai"
  },
  {
   "id": "technology.computing.ai.perception_language",
   "label": "Language and vision AI",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.perception_language"
  },
  {
   "id": "technology.computing.ai.computer_vision",
   "label": "Computer vision",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.computer_vision"
  },
  {
   "id": "technology.computing.ai.cv_methods",
   "label": "Vision methods and geometry",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.cv_methods"
  },
  {
   "id": "technology.computing.ai.cv_methods.low_level_analysis",
   "label": "Low-level image analysis",
   "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.cv_methods.low_level_analysis"
  }
 ],
 "geo": [
  {
   "id": "geo.nongeo.t2001.technology.computing.ai.perception_language",
   "label": "Non-geographic · 2001 to 2020: Language and vision AI",
   "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.perception_language",
   "path": [
    {
     "id": "geo.nongeo",
     "label": "Non-geographic",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo"
    },
    {
     "id": "geo.nongeo.t2001",
     "label": "Non-geographic · 2001 to 2020",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001"
    },
    {
     "id": "geo.nongeo.t2001.technology",
     "label": "Technology and the built world",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology"
    },
    {
     "id": "geo.nongeo.t2001.technology.computing",
     "label": "Computing and digital systems",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing"
    },
    {
     "id": "geo.nongeo.t2001.technology.computing.ai",
     "label": "Artificial intelligence and data",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai"
    },
    {
     "id": "geo.nongeo.t2001.technology.computing.ai.perception_language",
     "label": "Language and vision AI",
     "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.perception_language"
    }
   ]
  }
 ],
 "excerpt": "The super-resolution generative adversarial network (SRGAN) is a machine learning method that reconstructs photo-realistic high-resolution images from low-resolution inputs, introduced by Christian Ledig and colleagues in 2016.",
 "snippet": "The super-resolution generative adversarial network (SRGAN) is a machine learning method that reconstructs photo-realistic high-resolution images from low-resolution inputs, introduced by Christian Ledig and colleagues in 2016.",
 "node": "technology.computing.ai.cv_methods.low_level_analysis",
 "markdown": "# Super-resolution generative adversarial network\n\nThe super-resolution generative adversarial network (SRGAN) is a machine learning method that reconstructs high-resolution images from low-resolution inputs using a generative adversarial network trained with a perceptual loss, so that outputs look photo-realistic rather than merely close to the reference in pixel value. Its authors presented it as the first framework capable of inferring photo-realistic natural images for 4× upscaling factors.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup>\n\n| Key fact | Detail |\n| --- | --- |\n| Task | Single-image super-resolution, evaluated mainly at a 4× upscaling factor with bicubic-downsampled inputs<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> |\n| Objective | Perceptual loss: VGG19-based content loss plus an adversarial loss weighted by \\( 10^{-3} \\)<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> |\n| Generator | 16 residual blocks (the SRResNet design) with two trained sub-pixel convolution layers for upscaling<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> |\n| Training data | 350 thousand random ImageNet images, 96 × 96 HR crops, mini-batch of 16<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> |\n| Set5, 4× | SRGAN-VGG54: PSNR 29.40 dB, SSIM 0.8472, MOS 3.58; SRResNet-MSE: 32.05 dB, 0.9019, MOS 3.37<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> |\n| Main variant | ESRGAN (2018) replaces residual blocks with Residual-in-Residual Dense Blocks and uses a relativistic discriminator<sup>[2](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)</sup> |\n| Practical successor | Real-ESRGAN trains on purely synthetic high-order degradations for real-world photos<sup>[3](https://arxiv.org/abs/2107.10833)</sup> |\n\n## How it works\n\nSRGAN trains two networks against each other. A generator \\( G_{\\theta_{G}} \\) maps a low-resolution image \\( I^{LR} \\) to a super-resolved image, and a discriminator \\( D_{\\theta_{D}} \\) is trained to differentiate between the super-resolved images and original photo-realistic images.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> The two solve the adversarial min-max problem\n\n\\[ \\min_{\\theta_{G}} \\max_{\\theta_{D}} \\; \\mathbb{E}_{I^{HR} \\sim p_{train}(I^{HR})}[\\log D_{\\theta_{D}}(I^{HR})] + \\mathbb{E}_{I^{LR} \\sim p_{G}(I^{LR})}[\\log(1 - D_{\\theta_{D}}(G_{\\theta_{G}}(I^{LR})))] \\]\n\nfollowing Goodfellow et al.'s GAN formulation; for better gradient behavior the generator minimizes \\( -\\log D_{\\theta_{D}}(G_{\\theta_{G}}(I^{LR})) \\) instead of \\( \\log[1 - D_{\\theta_{D}}(G_{\\theta_{G}}(I^{LR}))] \\).<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> The adversarial term pushes solutions toward the natural image manifold, which is what produces textured, photo-realistic output that a pixel-wise mean-squared-error objective smooths away.\n\nThe full perceptual loss combines this adversarial term with a content loss:\n\n\\[ l^{SR} = \\underbrace{l^{SR}_{X}}_{\\text{content loss}} + \\underbrace{10^{-3} \\, l^{SR}_{Gen}}_{\\text{adversarial loss}} \\]\n\nThe content loss is the squared [Euclidean distance](https://www.edgepedia.org/euclidean-distance) between feature representations of the reconstructed image \\( G(I^{LR}) \\) and the reference \\( I^{HR} \\), taken from ReLU activation layers of a pre-trained 19-layer VGG network; feature maps are rescaled by a factor of 1/12.75 so VGG losses are comparable in scale to the MSE loss.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> The authors found the VGG/5.4 content loss to yield the perceptually most convincing results.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup>\n\nArchitecturally, the generator is the SRResNet design: \\( B = 16 \\) residual blocks, each with two 3 × 3 convolutional layers with 64 feature maps followed by batch normalization and ParametricReLU, and resolution is increased by two trained sub-pixel convolution layers. The discriminator follows Radford et al.'s guidelines, using LeakyReLU activation (\\( \\alpha = 0.2 \\)), avoiding max-pooling, and stacking eight 3 × 3 convolutional layers whose filter count increases by a factor of 2 from 64 to 512, with strided convolutions, two dense layers, and a final sigmoid.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup>\n\n## How it is done\n\nA practitioner trains in this order. First, low-resolution training pairs are created by bicubic downsampling with factor \\( r = 4 \\). The original SRGAN was trained on a random sample of 350 thousand images from ImageNet on an NVIDIA Tesla M40 GPU; for each mini-batch, 16 random 96 × 96 HR sub-images of distinct training images are cropped. Low-resolution inputs are scaled to [0, 1] and HR images to [-1, 1]. Optimization uses Adam with \\( \\beta_{1} = 0.9 \\), a learning rate of \\( 10^{-4} \\) for \\( 10^{5} \\) iterations followed by \\( 10^{-5} \\) for another \\( 10^{5} \\) iterations; the implementation used Theano/Lasagne.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup>\n\nA key step is initialization: the generator is first trained as an MSE-based SRResNet, and that network serves as the initialization for GAN training to avoid undesired local optima.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> A 2024 benchmark study confirms the practice, pretraining the generator with L1 loss only because training from scratch with adversarial loss yields artifacts.<sup>[4](https://arxiv.org/html/2405.17261v2)</sup>\n\nThe ESRGAN line follows the same recipe with updated settings: ×4 bicubic downsampling, mini-batch 16, 128 × 128 HR patches, and PSNR-oriented pretraining with L1 loss before perceptual training.<sup>[2](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)</sup> [Evaluation](https://www.edgepedia.org/evaluation) runs the trained model on standard test sets and reports PSNR, SSIM, and mean opinion score (MOS).\n\n## Origin\n\nSRGAN was reported in the paper \"Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network\" by Christian Ledig and colleagues, which first appeared on arXiv in 2016.<sup>[5](https://doi.org/10.17863/cam.51996)</sup> The paper itself states that, to the authors' knowledge, it is the first framework capable of inferring photo-realistic natural images for 4× upscaling factors, and a MOS test on three public benchmark datasets confirmed it as state of the art for photo-realistic SR at 4× by a large margin.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> The same paper introduced SRResNet, a 16-block residual network optimized for MSE that set a new state of the art in PSNR and SSIM at 4×; SRResNet serves as both a precursor and the generator backbone.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> The method builds on earlier work it cites directly: the GAN min-max formulation, the sub-pixel convolution layers for resolution increase, the DCGAN architectural guidelines, and the 19-layer VGG network used for the content loss.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup>\n\n## Variants\n\n**ESRGAN** (Xintao Wang and colleagues, 2018, arXiv) makes three changes: it replaces the residual block with a Residual-in-Residual Dense Block (RRDB) without batch normalization, combining multi-level residual and dense connections; it borrows the relativistic GAN idea so the discriminator predicts relative realness instead of the absolute probability that an input is real; and it computes the perceptual loss on features before activation, because activated features after a very deep network are very sparse.<sup>[2](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)</sup><sup> • </sup><sup>[6](https://doi.org/10.48550/arxiv.1809.00219)</sup> ESRGAN won first place in the PIRM2018-SR Challenge (region 3) with the best perceptual index.<sup>[2](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)</sup>\n\n**Real-ESRGAN** extends ESRGAN to practical restoration with pure synthetic data, using a high-order degradation process to simulate complex real-world degradations, including sinc filters for ringing and overshoot artifacts. It replaces the VGG-style discriminator with a U-Net discriminator with spectral normalization that outputs per-pixel realness values for detailed gradient feedback, and trains in two stages: a PSNR-oriented Real-ESRNet (L1 loss) initializes the generator, then [Real-ESRGAN](https://www.edgepedia.org/real-esrgan) trains with L1, perceptual, and GAN losses weighted {1, 1, 0.1}.<sup>[3](https://arxiv.org/abs/2107.10833)</sup>\n\nThe wider successor family improves different components: RankSRGAN (Wenlong Zhang and colleagues, 2019, arXiv) optimizes perceptual quality with a ranker,<sup>[7](https://doi.org/10.48550/arxiv.1908.06382)</sup> and later work has advanced architecture, training, discriminators, and data preprocessing across the GAN-SR lineage.<sup>[4](https://arxiv.org/html/2405.17261v2)</sup>\n\n## Applications\n\nSRGAN's headline result is the accuracy-versus-perceptual-quality trade-off. On Set5 at 4×, SRGAN with the VGG54 content loss reaches PSNR 29.40 dB, SSIM 0.8472, and MOS 3.58, while MSE-trained SRResNet reaches PSNR 32.05 dB, SSIM 0.9019 but a lower MOS of 3.37.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> On BSD100, SRGAN's PSNR of 25.16 dB falls below bicubic interpolation (25.94 dB) and SRCNN (26.68 dB), yet its MOS of 3.56 far exceeds bicubic (1.47) and SRCNN (1.87).<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> PSNR-oriented methods such as SRCNN therefore win on pixel fidelity while losing badly on perceived realism.\n\n## Limitations and alternatives\n\n**Hallucination and artifacts.** ESRGAN's authors describe SRGAN as a seminal work capable of generating realistic textures whose hallucinated details are often accompanied by unpleasant artifacts.<sup>[2](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)</sup> On complex textures such as grass and branches, comparative results on BSDS100 show SRGAN producing distortion and noise where higher-PSNR algorithms retain image quality.<sup>[8](https://www.mdpi.com/2073-8994/12/3/449)</sup> Real-ESRGAN's authors list twisted lines from aliasing, GAN training artifacts on some samples, and inability to remove out-of-distribution complicated degradations as limitations.<sup>[3](https://arxiv.org/abs/2107.10833)</sup>\n\n**Training instability.** In the original training, after only 20 thousand iterations the generator substantially diverged from its SRResNet initialization and produced reconstructions with a lot of high-frequency content, including noise; the authors also note that deeper SRGAN variants are increasingly difficult to train due to high-frequency artifacts.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup> A survey of diffusion-based SR lists GAN-based methods as susceptible to mode collapse, with a sizeable computational footprint, occasional failure to converge, and a need for additional stabilization methods.<sup>[9](https://ar5iv.labs.arxiv.org/html/2401.00736)</sup> By contrast, the 2024 benchmark study reports no optimization difficulties: after several hundreds of iterations its GAN models achieved generator-discriminator equilibrium and improved steadily until convergence.<sup>[4](https://arxiv.org/html/2405.17261v2)</sup> L1 or MSE pretraining of the generator is the common stabilizer across these accounts.<sup>[1](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)</sup><sup> • </sup><sup>[4](https://arxiv.org/html/2405.17261v2)</sup>\n\n**Alternatives.** PSNR-oriented networks (SRResNet, SRCNN) trade realism for pixel accuracy. Diffusion-based super-resolution has challenged GANs: human raters perceive SR images generated by diffusion models as more realistic than those generated by GANs, and StableSR uses a time-aware encoder trained in tandem with a frozen [Stable Diffusion](https://www.edgepedia.org/stable-diffusion) latent diffusion model.<sup>[9](https://ar5iv.labs.arxiv.org/html/2401.00736)</sup> Yet a 2024 head-to-head benchmark finds GANs remain competitive with diffusion methods.<sup>[4](https://arxiv.org/html/2405.17261v2)</sup>\n\n## References\n\n1. [Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (CVPR 2017)](https://openaccess.thecvf.com/content_cvpr_2017/papers/Ledig_Photo-Realistic_Single_Image_CVPR_2017_paper.pdf)\n2. [ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (ECCVW 2018)](https://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf)\n3. [Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data](https://arxiv.org/abs/2107.10833)\n4. [Does Diffusion Beat GAN in Image Super Resolution?](https://arxiv.org/html/2405.17261v2)\n5. [Ledig, Christian and colleagues (2016). Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network. arXiv (Cornell University).](https://doi.org/10.17863/cam.51996)\n6. [Wang, Xintao and colleagues (2018). ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1809.00219)\n7. [Zhang, Wenlong and colleagues (2019). RankSRGAN: Generative Adversarial Networks with Ranker for Image Super-Resolution. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1908.06382)\n8. [Generative Adversarial Network-Based Super-Resolution Considering Quantitative and Perceptual Quality](https://www.mdpi.com/2073-8994/12/3/449)\n9. [Diffusion Models, Image Super-Resolution And Everything: A Survey](https://ar5iv.labs.arxiv.org/html/2401.00736)\n\n---\n*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Artificial intelligence and data › Language and vision AI › Computer vision › Vision methods and geometry › Low-level image analysis*\n\n*Initially written Sep 29, 2026 · Edited: Oct 15, 2026*\n\n*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*\n\nLicense: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license\n",
 "same_as": [],
 "url": "https://www.edgepedia.org/super-resolution-generative-adversarial-network",
 "markdown_url": "https://www.edgepedia.org/super-resolution-generative-adversarial-network.md",
 "license": {
  "name": "Edgepedia Community License 1.0",
  "url": "https://www.edgepedia.org/edgepedia/license",
  "summary": "Free with credit, commercial use included. AI training is open to everyone. For other uses, organizations over USD 100M in revenue or 100M monthly users license separately.",
  "spdx": "LicenseRef-Edgepedia-Community-1.0"
 },
 "credit": "\"Super-resolution generative adversarial network\", Edgepedia (EdgeChat), https://www.edgepedia.org/super-resolution-generative-adversarial-network. Edgepedia Community License 1.0.",
 "credit_md": "\"[Super-resolution generative adversarial network](https://www.edgepedia.org/super-resolution-generative-adversarial-network)\", Edgepedia (EdgeChat), [https://www.edgepedia.org/super-resolution-generative-adversarial-network](https://www.edgepedia.org/super-resolution-generative-adversarial-network). [Edgepedia Community License 1.0](https://www.edgepedia.org/edgepedia/license).",
 "credit_html": "\"<a href=\"https://www.edgepedia.org/super-resolution-generative-adversarial-network\">Super-resolution generative adversarial network</a>\", Edgepedia (EdgeChat), <a href=\"https://www.edgepedia.org/super-resolution-generative-adversarial-network\">https://www.edgepedia.org/super-resolution-generative-adversarial-network</a>. <a href=\"https://www.edgepedia.org/edgepedia/license\">Edgepedia Community License 1.0</a>.",
 "speakable": "The super-resolution generative adversarial network is a machine learning method that reconstructs photo-realistic high-resolution images from low-resolution inputs, introduced by Christian Ledig and colleagues in 2016."
}