GENERATIVE ADVERSARIAL NETWORKS FOR SINGLE IMAGE SUPER RESOLUTION AND ENHANCEMENT.
A generative adversarial network is also known as GAN, which is a part of deep learning introduced by Ian Goodfellow and his team in 2014. GAN network consists of 2 networks. First thing is Generative algorithm and second thing is Discriminative algorithm. Generative is used to get the desire output...
| Publicado en: | Turkish Journal of Physiotherapy Rehabilitation Vol. 32; no. 2; pp. 2075 - 2087 |
|---|---|
| Autores principales: | , , |
| Formato: | equations & formulas pictorial tables/charts Journal Article |
| Publicado: |
Turkish Journal of Physiotherapy & Rehabilitation
2021
|
| Acceso en línea: | Ver este registro en EBSCOhost |
| Sumario: | A generative adversarial network is also known as GAN, which is a part of deep learning introduced by Ian Goodfellow and his team in 2014. GAN network consists of 2 networks. First thing is Generative algorithm and second thing is Discriminative algorithm. Generative is used to get the desire output and Discriminative algorithm is used to classify the ground truth and predicted output of the Generator network. There are many types of GAN networks are avaliable such as CycleGan, InfoGan and SRGAN etc. While zooming deeper into an image it will become blurry even you were taken that image by DSLR camera. Interpolation technique is used to solve this problem. But interpolation leads to in-accurate in pixels. To overcome this disadvantages we are going to implement learning based algorithm called SRGAN. This SRGAN consists two multilayer different algorithms called generator network is just as deconvolution network consists of many residual block, which were taken from Residual Network (ResNet) and another network is discriminator network whose architecture were taken from VGG-19. In this project our aim is to train the generator network to get the output of the high resolution image. The duty of the Discriminator network to classify the ground truth image (actual image) or output image (generated image by generator). The goal of the generator is try to fools the discrimintor to misclassify the fake image instead of real image. Discriminator network tries to point out the generated image as fake image. The Competition between these two network leads to good accuracy can be able to produce high quality image. |
|---|