AI-powered software a game-changer for 3D graphics

The consumption of 3D content is facing exponential growth. As the supply of 3D artists and content is not meeting demand, something has got to give, and industries are crying out for disruption. One EU project created the world’s first AI powered ArtEngine automating up to 80-90 % of the time-consuming and expensive parts of a 3D artist’s scan-based workflow.
Read the post in French | Lire l'article en Français 

The demand for 3D content now goes far beyond the gaming and VR/VFX industries. Fashion, furniture, aerospace, automotive, industrial design, architecture, and many other sectors are crying out for more 3D content. To keep up with growing trends, artists are moving to scan-based workflows; scanning the world around us and importing it into the virtual world.

Whether it’s building a virtual world creating products or designing fabrics the process is time-consuming and expensive. Studios creating 3D content are already at capacity failing to meet the current market demands and prior to Artomatix had no ability to meet the projected future rise.

Artomatix, the company behind the EU-supported ArtEngine project, has come up with a solution using Artificial Intelligence (AI) and neural networks. It can take the artist's scans and convert them into production-ready, photo-realistic materials. Everyday activities for an artist such as seam removal, expanding materials to much larger sizes, removing unwanted features or artifact compression (that currently make up 80-90 % of the time-consuming tasks) are now automated.

“Tasks that once took hours and days, now take seconds and minutes. Artists now have their time freed up to work on more value-added tasks. This is a game-changer for the 3D content creation!” explains Dr. Eric Risser, Founder and CTO at Artomatix

Advances in Convolutional Neural Network Image-to-Image translation inspired Artomatix to pursue several highly valuable features related to image enhancement or ‘upgrading’. “This feature suite takes low-resolution input images and produces high-resolution outputs by hallucinating new fine-scale details not existing in the original image,” explains Dr Risser.

Using neural networks, ArtEngine can imagine details based on the context provided by the low-resolution image. Powered by this novel and patent pending network strategy, are a suite of related features such as the ability to take heavily damaged images due to compression and restore to their former quality, with noise removal and deblurring also included.

Another solution the project has developed is Automatic Texture Segmentation. In the scanning process, artists can spend a significant amount of their time masking unwanted features or dividing a heterogeneous scan into its set of individual homogeneous parts. Artomatix has lightened this burden by using AI to automatically recognise and identify the different textural components from a scan. The core technology for this feature has been developed and a prototype has been built.

Artists welcoming the new tools

A big shift for Artomatix, over the past two years, is seeing its target customer base shift, they are now projecting 40-50 % of their customer base to be from outside of the gaming and entertainment space. “With the introduction of Artomatix into all mainstream industries, Artists can now complete 90% of their tasks in 5 % of the time; with Artists seeing 10X productivity gains. The artist productivity gains have put Artomatix on the world stage,” adds Joe Blake, CEO at Artomatix. Independent analysis supports this claim.

Blake says, “Looking back over 2018, in particular, we are delighted to be one of the very few companies to ever secure Horizon 2020 for a second time. We were humbled and excited to be selected as one of the first companies for the Irish Government’s DTIF funding.” Artomatix also secured its first commercial deals this year.

Source: Cordis

Illustration Photo: 3D modelling (credits: Nathanael Coyne / Flickr Creative Commons Attribution-NonCommercial-NoDerivs 2.0 Generic (CC BY-NC-ND 2.0))

Read more

Comments

No comments to display.