Via Fast Company:
Like the work that came out of the golden age of Dutch painting, these paintings may look like the product of oil, brushes, and fingers. But, like the rest of Albiac’s work, these portraits are actually the result of the artist’s computer code.
This series, like many other similar projects, was created using a Generative Adversarial Network that Albiac developed himself. The system consists of two neural networks. One is trained to generate portraits based on training data from a database of real Dutch masterpieces, while the second one evaluates its output and compares it to the actual paintings. Albiac told me via email that he used around 300 paintings to create the training data, selected from the high-res digital collection published by the Rijksmuseum: “Most portraits in my training dataset are 17th century Dutch paintings, some other portraits from 18th and 19th centuries.” With that evaluation data in hand, the first AI refined its skills, producing new paintings that were evaluated again by the second neural network in an interactive process that resulted in the images you see in this article.
Every Tuesday is Art Tuesday here at Adafruit! Today we celebrate artists and makers from around the world who are designing innovative and creative works using technology, science, electronics and more. You can start your own career as an artist today with Adafruit’s conductive paints, art-related electronics kits, LEDs, wearables, 3D printers and more! Make your most imaginative designs come to life with our helpful tutorials from the Adafruit Learning System. And don’t forget to check in every Art Tuesday for more artistic inspiration here on the Adafruit Blog!