Example results on several image restoration problems. We use deep neural networks, but we never train/pretrain them using datasets. We use them as a structured image prior. Deep convolutional networks have become a popular tool for image generation and restoration. Generally, their excellent performance is imputed to their ability to learn realistic image priors from a large number of example ima
![Deep Image Prior](https://cdn-ak-scissors.b.st-hatena.com/image/square/5c27e26b1ed501cea34fa44c33452fea2f35dcd1/height=288;version=1;width=512/https%3A%2F%2Fraw.githubusercontent.com%2FDmitryUlyanov%2Fdeep-image-prior%2Fmaster%2Fdata%2Fteaser_compiled.jpg)