ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks
Citations
837 citations
Cites background or methods from "ESRGAN: Enhanced Super-Resolution G..."
...Some other models [32], [103], [147] also adopt this experience and achieve performance improvements....
[...]
...In practice, researchers often combine multiple loss functions by weighted average [8], [25], [27], [46], [141] for constraining different aspects of the generation process, especially for distortion-perception tradeoff [25], [103], [142], [143], [144]....
[...]
...[103], [155] propose a network interpolation strategy....
[...]
...And the ESRGAN [103] employs relativistic GAN [134] to predict the probability that real images are relatively more realistic than fake ones, instead of the probability that input images are real or fake, and thus guide recovering more detailed textures....
[...]
...Thus it produces visually more perceptible results and is also widely used in this field [8], [25], [29], [30], [46], [103], where the VGG [128] and ResNet [96] are the most commonly used pre-trained CNNs....
[...]
697 citations
Cites background from "ESRGAN: Enhanced Super-Resolution G..."
...Generative adversarial networks (GANs) [9] have shown impressive results in many computer vision tasks such as image synthesis [3, 25, 7], colorization [15, 36] and superresolution [21, 34]....
[...]
581 citations
Cites methods from "ESRGAN: Enhanced Super-Resolution G..."
...[22] used Gram matrix based texture matching loss to enforce local similar textures, while ESRGAN [32] enhanced SRGAN by introducing RRDB with relativistic adversarial loss....
[...]
...The SISR methods include SRCNN [3], MDSR [17], RDN [40], RCAN [39], SRGAN [16], ENet [22], ESRGAN [32], RSRGAN [38], among which RCAN has achieved state-of-the-art performance on both PSNR and SSIM in recent years....
[...]
...Sajjadi et al. [22] used Gram matrix based texture matching loss to enforce local similar textures, while ESRGAN [32] enhanced SRGAN by introducing RRDB with relativistic adversarial loss....
[...]
550 citations
428 citations
References
123,388 citations
111,197 citations
"ESRGAN: Enhanced Super-Resolution G..." refers methods in this paper
...For optimization, we use Adam [39] with β1 = 0....
[...]
...For optimization, we use Adam [39] with β1 = 0.9, β2 = 0.999....
[...]
55,235 citations
"ESRGAN: Enhanced Super-Resolution G..." refers methods in this paper
...3 We use pre-trained 19-layer VGG network[37], where 54 indicates features obtained by the 4 convolution before the 5 maxpooling layer, representing high-level features and similarly, 22 represents low-level features....
[...]
49,914 citations
38,211 citations
"ESRGAN: Enhanced Super-Resolution G..." refers background in this paper
...eature transform to eectively incorporate semantic prior in an image and improve the recovered textures. Throughout the literature, photo-realism is usually attained by adversarial training with GAN [15]. Recently there are a bunch of works that focus on developing more eective GAN frameworks. WGAN [31] proposes to minimize a reasonable and ecient approximation of Wasserstein distance and regulariz...
[...]
...mprove the visual quality of SR results. For instance, perceptual loss [13,14] is proposed to optimize super-resolution model in a feature space instead of pixel space. Generative adversarial network [15] is introduced to SR by [1,16] to encourage the network to favor solutions that look more like natural images. The semantic image prior is further incorporated to improve recovered texture details [17...
[...]