Vincent™ AI Artist: Generative Adversarial Networks (GAN) as seen in NVIDIA GTC keynote

See our method in motion here in NVIDIA CEO Jensen Huang’s GTC 2017 Europe keynote speech and at GTC 2018 bash

A whole of eight neural networks blended in adversarial loops acquired to:
– Reproduce eight,000 training illustrations or photos of Western artwork
– Reproduce final information these types of as brush marks and canvas texture
– Extract edges from photos in a handy way for the software
– Change in between two representations of an image: Yellow outlines and the comprehensive image

As soon as qualified, the Vincent is equipped to infer opportunity artworks from consumer-drawn outlines in actual time.

The era of megapixel illustrations or photos centered only on a training established of just eight,000 illustrations or photos is special at the time of publishing.