Hidden behind Leonardo da Vinci's, 1495-1508, Virgin of the Rocks, hides the traces of an earlier work; a sketch of an alternative scene of The Virgin and Christ Child in a mysterious rocky landscape. This work explores the possible set of paintings of what could have been da Vinci’s Madonna.
Using Leonardeschi paintings we trained a GAN based model to map between edgemaps - representative of x-rays - and Leonardeschi paintings. We modified and co-registered the x-ray of da Vinci's Virgin on the Rocks, from Imperial College London, with the x-ray trace produced by the National Gallery, where the Virgin of the Rocks is currently held.
We then created a set of augmented segments of the co-registered image, zoomed in on The Virgin, that vary in brightness, contrast, and sharpness. These augmented segments were then passed to the trained Leonardeschi model. Running the model enabled transformation of the augmented segments into a set of paintings, producing the work shown.
This piece featured in NVIDIA’s Fall 2020 GTC (GPU Technology Conference) Inaugural AI Art Gallery. Further details of which are shown in Cann et al. (2021), Resolution enhancement in the recovery of underdrawings via style transfer by generative adversarial deep neural networks, Computer Vision and Art Analysis, IS&T, Springfield, VA, 2021.