The paper presents SRGAN, a generative adversarial network (GAN) for image super-resolution (SR) that aims to recover photo-realistic textures at large upscaling factors (4×). Unlike traditional methods that focus on minimizing mean squared error (MSE), SRGAN introduces a perceptual loss function, which includes both an adversarial loss and a content loss. The adversarial loss encourages the super-resolved images to be indistinguishable from natural images, while the content loss, inspired by perceptual similarity, ensures that the high-frequency details are preserved. The deep residual network (ResNet) used in SRGAN is trained to produce high-quality reconstructions, and extensive mean opinion score (MOS) tests on public benchmark datasets demonstrate that SRGAN significantly outperforms state-of-the-art methods in terms of perceptual quality. The results show that SRGAN's reconstructions are closer to the original high-resolution images than those obtained with other methods, highlighting its superior performance in recovering fine textures and overall visual quality.The paper presents SRGAN, a generative adversarial network (GAN) for image super-resolution (SR) that aims to recover photo-realistic textures at large upscaling factors (4×). Unlike traditional methods that focus on minimizing mean squared error (MSE), SRGAN introduces a perceptual loss function, which includes both an adversarial loss and a content loss. The adversarial loss encourages the super-resolved images to be indistinguishable from natural images, while the content loss, inspired by perceptual similarity, ensures that the high-frequency details are preserved. The deep residual network (ResNet) used in SRGAN is trained to produce high-quality reconstructions, and extensive mean opinion score (MOS) tests on public benchmark datasets demonstrate that SRGAN significantly outperforms state-of-the-art methods in terms of perceptual quality. The results show that SRGAN's reconstructions are closer to the original high-resolution images than those obtained with other methods, highlighting its superior performance in recovering fine textures and overall visual quality.