At the GPC 2019 conference, NVIDIA introduced AI Playground — a smart platform for photo editing based on the neural networks. The tool is intuitive and made to cover the most popular designer’s tasks.
The project pushes deep learning closer to the real needs of users and designers, and the first 3 products of the AI Playground — image imprinting, artistic style transfer, and photorealistic image synthesis — seem to prove it well. And put lots of fun into casual chores, basically.
Image Inpainting. The technology of a convolutional neural network enables up-to-date image reconstruction, where you can fill a selected area with new pixels, independent from the erased fragments. It means you can remove a pass-by from a group photo or a suspiciously-looking strand from a selfie quite easily
Artistic Style Transfer. The name has already told the story. Here the NVIDIA team means to help you copy the style of a given image without downloading and purchasing any filters, actions, and effects. That sounds practical, and I hope the idea will get its further development. The program should go live soon, and the team promises infinite possibilities — so I recommend to subscribe for the project updates, we just can’t miss it.
Photorealistic Image Synthesis. You may be familiar with pix2pix introduced by Christopher Hesse a couple of years ago. It was the time where everyone drew kitties, men, buildings, shoes or bags, and let the neural networks transform them into realistic (errr, no) images.
NVIDIA promises to attain speed, resolution, realism, complexity, and editability by evolving the GAN, a generative adversarial network. The image synthesis program is also under construction but can go live at any moment, and I’m super excited to see how it will works for myself since I hope the images will go truly realistic this time.