We present SinGAN, an unconditional generative model that can be gained from a single natural image. 



Our Model is  prepared to catch the inward conveyance of patches inside the image, and is then ready to produce high caliber, differing tests that convey a similar visual substance as the image. SinGAN contains a pyramid of completely convolutional GANs, each answerable for learning the fix dissemination at an alternate size of the image. This permits creating new examples of arbitrary size and angle proportion, that have critical inconstancy, yet keep up both the worldwide structure and the fine surfaces of the preparation image. Rather than past single image GAN plans, our methodology isn't constrained to surface images, and isn't contingent (for example it produces tests from commotion). Client studies affirm that the produced tests are ordinarily confounded to be genuine images. We outline the utility of SinGAN in a wide scope of image control errands.

Project Paper Click Here

Source Code: Click Here