FEATURED COMPUTER VISION
Native AI Learns to See and Understand Language
The researchers developed a new type of computer model called NEO, which can understand images and text together in one system,.
Read Full Paper →AI Edits Images from Any Angle with Ease
We've developed a new way to combine multiple images taken from different angles into a single edited image that looks realistic and consistent. This method could be useful in many fields, such as photography or filmmaking, where being able to accurately edit and combine multiple views of an object or scene can help create more convincing and detailed visuals.
Read Paper →AI Learns to Build Better by Imitating Human Design Skills
Scientists created a virtual game world where machines can be built from standardized parts, to see if large language models can learn to create complex systems. If successful, this could lead to more intelligent machines that can be used in engineering and other fields, which could have significant benefits for society.
Read Paper →
AI Learns Image Editing from Instructions Alone
Scientists created a new way to teach computers to edit images using written instructions without needing large amounts of specific example data. This method could make it easier and more efficient for computers to learn how to edit images, potentially leading to better image editing tools in the future.
Read Paper →AI Learns from Body Language to Create Realistic Interactions
Scientists analyzed a large collection of videos showing people interacting closely with each other, looking for patterns in how their bodies move and position themselves. They developed a new system called Ponimator that can use these patterns to generate realistic animations of people interacting, which could be useful in virtual reality games or training simulations where lifelike interactions are important.
Read Paper →