The Eye of AI #3

Computer vision is a truly transformational technology for the retail industry, but it has applications that touch nearly every aspect of our lives. This is our weekly summary of what happens when cameras, computers and great ideas get together.

Computer Vision Caring for Human Vision

UC San Diego's electrical engineering students and ophthalmologists at the Jacobs Retina Center have teamed up to tackle eye diseases. Their partnership has led to some impressive results, like an AI tool that can diagnose eye diseases with 80% accuracy just from images. They've also developed methods to track changes in the retina over time, such as due to diabetic macular degeneration. This unique collaboration blends engineering and medical expertise to improve patient care and highlight the ethical considerations of using AI in healthcare.

Automated Litter Picking

Did you ever try to make the ratatouille from Ratatouille? Fun fact - that dish was actually a confit byaldi rather than a real ratatouille. Anyway, over at the Istituto Italiano di Tecnologia's Dynamic Legged Systems lab, they have built their own version of the adorable WALL-E (Waste Allocation Load Lifter - Earth Class). Their version, called Vero - a Vacuum-cleaner Equipped quadraped RObot, has been designed to clean the cigarette butts off local parks and beaches. It's a great idea and has enormous benefits for beach goers and the associated wildlife, but I think we can all agree that these dog-shaped robots are just a little bit scary.

Safer Streets City-Wide

Do you know the way to San Jose? And when you get there, do you know whether the streets are safe? Well, the Toyota Mobility Foundation has just awarded San José, California, a $260,000 grant to use AI and computer vision to improve traffic safety. The project will help detect road hazards like illegal parking and abandoned mattresses that endanger cyclists and pedestrians. This initiative aims to make roads safer by quickly identifying and addressing issues, allowing city staff to focus on solving problems.

Monitoring Tropical Rivers Through Cloud

When you look at a rainbow, that thin curve of colour marks the limits of our ability to perceive the world (??), but the sun's energy goes way above and below what we can see. Climate monitoring satellites tend to use optical sensors to monitor the Earth's changing patterns but, just like our eyes, they miss a lot of useful information. So scientists at UC Berkley are using computer vision based on L-Band microwaves to watch the movement of water in tropical rivers that are often covered by cloud (and rain) to predict flood threats and model greenhouse gas emissions in tropical wetlands.

A Brief Geek-Out

'She's a model and she's looking good' droned Kraftwerk in 1978. Here in 2024 we're more interested in models that are good at looking. So here's a whirl around the latest research:

• Engineers at Google Deep Mind and the University of Washington (Go huskies!) have teamed up to reduce the processing requirements of computer vision applications by developing MoNE (Mixture of Nested Experts) to effectively deal with the video redundancy
• Meta released SAM2 - a new version of its Segment Anything Model for image and video processing as an open-source project. Immediate applications are genAI or video creation based, but there is potential to speed up the annotation of visual data.
• More work from Meta - this blending image and text processing capabilities into a multimodal mixture of experts they're calling MoMa. It's still early days for large-scale multimodal systems (ones that can deal with more than one kind of input and output data), but this looks like a significant step

That's everything for this week. Please keep an eye on the SAI Group blog for everything that we're talking about, including a new article this week about the EU AI Act.

Got some cool tech to share? Whether its your own project, or just something you saw and thought "I want people to know about this!", let us know about it and we'll include it in upcoming editions.