All articles tagged with #computer vision
A computer-vision and robotics expert built a DIY AI system that uses a camera and deep learning to detect mosquitoes and a laser to kill them, with safeguards to cut power if people or flammable materials appear in view. The designer said all mosquitoes in his home were eliminated after one night, and the piece also notes CDC pest-control tips and non-laser alternatives for indoor mosquitoes.
UW researchers embedded low‑resolution, black‑and‑white cameras into Sony WF‑1000XM3 wireless earbuds (dubbed VueBuds) to give them vision and run AI-driven scene understanding, echoing rumors about camera-enabled AirPods. The project aims for AI-assisted tasks like navigation and environment-aware reminders, but faces battery life and privacy tradeoffs, plus reliance on cloud processing. Early results show responsive performance comparable to Ray‑Ban/Meta AI glasses, with the potential for a more refined version from big‑tech players in the future.
Gary Marcus argues that ChatGPT’s new image engine is visually impressive but does not demonstrate true understanding. He points to labeling errors in bike diagrams and odd results from a custom tandem-bike prompt as evidence that the system can imitate understanding without grasping how parts function. The piece emphasizes that regurgitating images isn’t the same as real comprehension in AI.
The piece explains how Tesla's FSD uses only camera input to create a live 3D representation of the world by fusing multi-view image features into a unified 3D space. Two patents—Vision-Based Occupancy Determination and Vision-Based Surface Determination—describe a voxel-based occupancy map and a 3D surface mesh, built from 2D images via transformers, temporal fusion, and deconvolution, enabling depth, motion, and material understanding without LiDAR. The two systems work together to inform prediction, path planning, and control, training on data from LIDAR/photogrammetry and camera feeds to produce a robust, dynamic world model for real-time driving decisions.
Bloomberg reports Apple plans to weave Visual Intelligence (computer vision) into a wave of upcoming devices—AirPods with cameras, first-gen smart glasses, and an AI pendant—to give gadgets “sight,” enable item recognition and navigation cues, and prompt actions. Critics argue the vision tech mirrors what’s already in Meta’s glasses, may be unreliable, and its everyday usefulness remains uncertain as rollout is not expected before late 2026 at the earliest.
Amazon is developing smart glasses for delivery drivers that use AI and computer vision to provide real-time navigation, hazard detection, and proof of delivery, aiming to make last-mile deliveries safer and more efficient while offering a hands-free experience.
Apple is close to acquiring Prompt AI's team and technology, including its flagship product Seemour, a security camera AI. The deal involves an acquihire strategy, with Prompt's employees being informed and offered roles at Apple, while the Seemour app will be discontinued. This move aligns with a broader trend of tech giants acquiring AI talent and technology to enhance their capabilities while avoiding regulatory scrutiny.
Apple is nearing a deal to acquire Prompt AI, a startup specializing in computer vision technology, as part of its strategic expansion into AI and machine learning capabilities.
A researcher has developed a low-cost air quality estimation system using a microcontroller with a camera and a trained AI model to analyze sky images, offering a novel approach that complements traditional sensors, though with some limitations in accuracy.
Velo AI has developed Copilot, a Raspberry Pi-powered bike light equipped with AI and machine learning to detect cars, alert cyclists about approaching vehicles, issue warnings to drivers, and record incidents. Priced at $400, the device aims to enhance cyclist safety and is being used in a partnership with Pittsburgh to analyze data for potential road safety improvements.
Researchers have created the largest-ever dataset of biological images, containing over 10 million images of plants, animals, and fungi, and developed a new vision-based artificial intelligence tool called BioCLIP to learn from it. The tool, which outperformed existing models, can classify images and discern species across the tree of life, making it useful for a wide range of biological studies. The AI model's ability to learn fine-tuned representations of images and its potential for unraveling biological mysteries make it a valuable tool for biologists.
A study by MIT found that artificial intelligence is currently too expensive to effectively replace the majority of jobs, with only 23% of tasks being cost-effective to automate using AI-assisted visual recognition. The study, funded by the MIT-IBM Watson AI Lab, examined over 1,000 visually-assisted tasks across 800 occupations and concluded that while the adoption of AI in industries like retail and healthcare is feasible, it is less so in areas like construction and real estate. The researchers suggested that the cost-benefit ratio of AI could improve by 2030 if data costs fall and accuracy improves, but concerns about AI's impact on jobs persist, with industry leaders cautioning against a recklessly fast AI rollout.
Apple has released a research paper detailing its generative AI technology called HUGS, which can create a digital human avatar from a short video in about 30 minutes. The technique, known as Human Gaussian Splats (HUGS), uses 3D Gaussian Splatting to create an animatable human within a scene. The system can disentangle the static scene and create a fully-animatable human avatar, filling in gaps for elements like cloth and hair. The process is about 100 times faster than other methods and has been trained in collaboration with the Max Planck Institute for Intelligent Systems. Apple has been exploring the concept of digital avatars for various applications, including FaceTime conversations and representing users in virtual environments.
Alexei Efros, a computer scientist at the Berkeley Artificial Intelligence Research Lab, discusses the history of computer vision for AI, his breakthroughs in the field, and the impact of his work on everyday life. He explains how his poor eyesight proved advantageous and emphasizes the importance of addressing human bias in contemporary AI.
Former Google Nest engineers have developed a new robot vacuum called Matic, which uses visual processing instead of spatial mapping to navigate. Equipped with five RGB cameras, Matic can see in real-time and avoid common obstacles like high-pile rugs and cables. It operates locally without a cloud component, ensuring data privacy. Matic can vacuum and mop, and its computer vision allows it to identify different floor types and switch between cleaning modes. It can respond to gesture commands and autonomously seek out dirty areas. The vacuum is designed to be quiet, has a one-liter capacity bin, and can handle wet spills. Matic is available for pre-order at a discounted price of $1,495 and is set to be delivered in March 2024.