MAGE: MIT’s Advanced System for Generating and Recognizing Images

MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a single system called Masked Generative Encoder (MAGE) that can handle both image recognition and generation tasks with high accuracy. Previously, building an image recognition or generation system required two separate processes, state-of-the-art generative modelling and self-supervised representation learning. However, MAGE does both by using a pre-training approach called masked token modelling. Researchers from MIT plan to perfect the technology further and expand its capabilities. It currently supports conditional image generation that allows users to specify criteria for the images they require. MAGE could have applications in industries like photo editing, visual effects and post-production.