Oxia Palus Ltd

Company description

Oxia Palus utilises AI and 3D printing to reconstruct famous artworks in order to display them in exhibitions.

Summary: Oxia Palus an artificial intelligence art collective. Our mission is to accelerate the adoption of artificial intelligence by artists and art conservators, so as to reconstruct the world's lost art and push the boundaries of creativity. Oxia Palus primarily produces lost artwork, by applying neural style transfer (NST) to modified x-radiographs of artwork with secondary interior artwork painted beneath a primary exterior painting. Extended: Our work is about applying a new technique in machine learning, called neural style transfer (NST), so as to create possible reconstructions of lost artwork. NST was developed by Gatys et al. in 2015, and can be used to create novel artistic work by rendering one image in the style of another. NST uses a previously trained convolutional neural network (CNN) called the VGG-19 network; a 19 layer neural network that has been trained on a large dataset of ImageNet images. Training on this large dataset of images allows the VGG-19 neural network to recognise low and high level features in images. A key part of the NST algorithm is defining and minimising the content and style cost functions (also known as error functions). The content cost function is an error between the content image and the generated image, similarly, the style cost function is an error between the style image and the generated image. Minimising the content cost function will make the generated image follow the style of the style image and minimising the style cost function will make the generated image follow the content of the content image. Once these functions are defined, we add them together to create a total cost function. We then use an optimisation algorithm to minimise the total cost function, which makes the generated image follow the content of the content image and the style of the style image simultaneously, which after several iterations creates stylised images. We use NST as a novel method of reconstructing lost artwork, by applying neural style transfer (NST) to modified x-radiographs of artwork with secondary interior artwork painted beneath a primary exterior painting. The significance of this work is that, to our knowledge, it's the first application of style transfer to x-radiographed imagery of artwork, known to have a hidden interior piece, with the aim to create a possible reconstruction of the underlying original. We believe the emergence of AI in art will significantly expand our creative horizon. We think that we're at the creative shoreline taking our first steps on land. In the field of art history, we believe AI will help by making it easier and cheaper to and reveal what lies beneath historical artwork. There are potentially thousands of underpaintings and underdrawings, each with their own story, that are hidden from world. This hidden art may hold important information about how art has developed throughout history.

Project impact

In September 2019 Oxia Palus featured in MIT Technology Review, https://www.technologyreview.com/s/614333/this-picasso-painting-had-never-been-seen-before-until-a-neural-network-painted-it/, for reconstructing a lost Picasso, La Femme Perdue and subsequently gained worldwide press coverage, for more details see: https://www.oxia-palus.com/press. This research titled 'Raiders of the Lost Art', arXiv: https://arxiv.org/pdf/1909.05677.pdf, was recently presented at NeurIPS 2019, the world's largest AI conference, in Vancouver and achieved 3rd in NVIDIA's Top 10 AI Developer Stories of 2019, https://news.developer.nvidia.com/nvidias-top-10-ai-developer-stories-of-2019/.