Paint the future: the neural network has learned to add non-existent details to a photo

Thanks to neural networks, graphic editorsthe future will be capable of impossible things today. For example, can you imagine that a regular computer program can independently draw a smile on a photograph of a person, remove glasses and change a hairstyle? It seems fantastic, but such a program has already been created and uploaded to GitHub, so any programmer can run it on his computer and edit his photos. The results of the program are impressive.

Based on the application liesgenerative-competitive neural network SC-FEGAN. Similar networks are used in many similar projects and consist of two parts. In this application, they are the Unet-like image generator and the SN-patchGAN discriminator, the first of which creates sample images, and the second decides whether they are suitable or not.

The program interface is quite simple - fromuser is required to upload a photo of a person, and draw on it new details. They can be extra strands of hair, a smile, and even a variety of jewelry. Moreover, the neural network is able to remove some objects like glasses, as well as change the color of the eyes and hair.

Unfortunately, at the moment run the programYou can only follow the instructions for ordinary users. However, do not despair - in the near future, the SC-FEGAN neural network can form the basis of browser-based or mobile applications for image editing.

Recently similar projects are far fromrarity. In mid-February, one of the developers of the Uber taxi service launched a website where every time a face of a nonexistent person is generated. You can read about the project and watch the video with the process of creating artificial faces in our material.

What other types of programs do you think neural networks can be embedded in? What new features would they have? You can dream up on this topic in the comments, or in our Telegram-chat!