Scientists from MIT and Adobe Research have developed a method for artificial intelligence (AI) to distinguish materials in images or videos, even when lighting or object shape affects the perception. The model was trained on 50,000 synthetic images of over 16,000 materials, but also performed well when presented with unrelated, real-life samples. The system can be tailored to particular materials by inputting a single pixel from it, and can help robots choose the necessary amount of force to work with an object made from that material. The research, which is due to be presented at the Siggraph 2023 conference, was led by Praofull Sharma.
/Reports, release notes, official announcements.