Articles tagged with Computer Vision
Thatchaphol Saranurak and Andrew Owens have been awarded Sloan Research Fellowships for their innovative work on graph networks and machine perception systems. Their research aims to create more efficient algorithms for computing dynamic systems, such as social networks and traffic patterns.
MIT researchers have introduced a new system called MiFly that enables drones to self-localize in indoor, dark, and low-visibility environments. The system uses radio frequency waves reflected by a single tag placed in the environment, allowing the drone to estimate its trajectory with high accuracy.
A new method developed by Osaka Metropolitan University accurately predicts housing prices in Osaka City, with neighborhood perception being a key factor. The approach achieves nearly 75% accuracy by combining existing property data with machine-learning-processed street view images.
The open-source AI model analyzes medical images, generates detailed reports, and answers clinical questions to streamline diagnostics and improve accuracy. BiomedGPT aims to democratize healthcare and reduce disparities amongst patients by providing easily accessible data to bolster underserved hospitals.
A comprehensive review of camouflaged object detection research highlights the potential of deep learning in recognizing objects in complex scenarios. The review analyzes traditional and deep learning approaches, emphasizing practical contributions and theoretical frameworks.
Researchers develop precision techniques using optical sensors and AI to facilitate efficient and accurate food drying. The study discusses three emerging smart drying techniques, providing practical information for the food industry.
A new tool developed by Penn State researchers uses computer vision and artificial intelligence to analyze placenta images, detecting abnormalities and risks such as neonatal sepsis. The PlacentaCLIP+ model has the potential to transform neonatal and maternal care in low- and high-resource settings.
Researchers from Bar-Ilan University discover that classifying objects together through Multi-Label Classification can yield better results than detecting individual objects. This new method allows networks to learn correlations between object combinations, making them more recognizable in real-life applications such as autonomous vehi...
The University of Tennessee Institute of Agriculture has won a four-year grant to create hands-on curriculum about AI-related technologies for future farmers and leaders. Selected students will test the curriculum in drones, robotics, and other smart agriculture technologies, gaining skills in coding, drone-work, and robotics.
Researchers develop a simple fix to an existing technique, enabling the generation of sharp, high-quality 3D shapes that rival top model-generated 2D images. The new approach improves upon previous methods by avoiding costly retraining and complex postprocessing.
Researchers at CAMERA have developed an open-source markerless motion capture system using computer vision and deep learning methods. The system estimates joint positions from regular 2D image data, providing unobtrusive analysis of body movements.
A study by Osaka University researchers found that visual landmarks can be difficult to find in certain environments, leading to motion sickness. They propose using radio-frequency localization, such as ultra-wideband sensing, to overcome these challenges and improve indoor augmented reality applications.
A new computational model called Multi-Stage Residual-BCR Net (m-rBCR) uses a unique frequency representation to solve deconvolution tasks with fewer parameters and faster processing times. The model demonstrates high performance on various microscopy datasets, outperforming traditional methods.
A new crowdsourcing system, FireLoc, uses a network of low-cost mobile phones to detect wildfires minutes—even seconds—after they ignite. The system prioritizes privacy and accurately maps wilderness fires to within 180 feet of their origin.
Researchers developed a novel AI approach to predict atomic-level chemical bonding information in 3D space, bypassing traditional supercomputer simulations. This methodology accelerates calculations by learning chemical bonding information using neural network algorithms from computer vision.
Researchers used facial recognition technology to track actor screen time in over 2,300 films, confirming a shift towards greater diversity. The study found that individual film casts are becoming more diverse, with non-leading roles exhibiting more variety than leading ones.
WorldScribe, a new software, uses generative AI to provide real-time text and audio descriptions of surroundings for people who are blind or have low vision. The tool can adjust the level of detail based on user commands or camera frame time.
A new method called Clio allows robots to make task-relevant decisions by identifying the parts of a scene that matter. In real experiments, Clio successfully mapped scenes at different levels of granularity based on natural-language prompts and enabled robots to grasp objects of interest.
Rice University researchers developed ElasticDiffusion, a method that separates local and global signals to create non-square aspect ratio images without visual imperfections. The new approach can improve consistency and realism in AI-generated images, but still requires significant computational power.
Researchers at Tsinghua University Press have developed BiRefNet, a bilateral reference framework that captures tiny-pixel features and achieves highly accurate high-resolution salient object detection and concealed object detection. The framework has numerous practical applications in various fields.
Researchers at the University of South Australia have developed an AI sensor that can accurately measure the orientation of the Milky Way in low light, using a technique inspired by the dung beetle. This system could improve navigation for drones and satellites in difficult lighting conditions.
Researchers at Jackson Laboratory have developed a non-intrusive method to accurately and continuously measure mouse body mass using computer vision. This approach reduces stress associated with traditional weighing techniques, improving data accuracy and reproducibility.
The Segment Anything Model has achieved significant breakthroughs in image segmentation, leveraging its data engine methodology and vast datasets. Researchers have proposed improvements and applications for the model, showcasing its versatility across various tasks and domains.
Researchers at Duke University have broken through the performance wall of adaptive radar systems using convolutional neural networks, paralleling computer vision. They've released a large open-source dataset for other AI researchers to build upon their work, aiming to tackle industry needs like object detection and tracking.
UCF's STRONG-AI initiative aims to uplift bright, low-income undergraduate students in pursuing well-rounded AI education through faculty and peer mentorship and scholarship. The program has received over 150 applications and will select 10-15 students annually based on financial aid eligibility and academic success.
A study found that large language models (LLMs) like ChatGPT underperform state-of-the-art detectors but can explain their analysis in plain language. LLMs' semantic knowledge makes them well-suited for detecting deepfakes, providing a common sense understanding of reality.
A new AI model developed by Surrey researchers and Stanford University can accurately identify objects in complex scene sketches, even from non-artists. The model achieved an 85% accuracy rate, outperforming previous approaches that relied on labelled pixels.
Researchers developed a technique called Multi-View Attentive Contextualization (MvACon) to improve AI's ability to map 3D spaces using 2D images from multiple cameras. MvACon significantly improved the performance of vision transformers in locating objects and detecting speed and orientation.
A new computer vision technique developed by MIT engineers significantly speeds up the characterization of newly synthesized electronic materials. The technique automatically analyzes images of printed semiconducting samples and quickly estimates two key electronic properties: band gap and stability.
A new study from the University of Tsukuba introduces an algorithm that determines the application ratio of various compression methods for minimizing data amount in CNNs. This leads to a 28 times smaller model and 76 times faster computation compared to previous models.
Researchers from Osaka University developed a mobile mixed reality (MR) system for intuitive flooding forecasts, allowing urban populations to view dynamic flood forecasts on their mobile devices. The system enables widespread participation in MR visualizations, improving community preparedness and response.
A novel approach to training AI systems uses information about spatial position to identify objects and navigate surroundings, inspired by children's visual development. The method improves contrastive learning models' effectiveness by incorporating simulated spatial context information, outperforming base models in various tasks.
Researchers developed a new method to enhance thermal image super-resolution by employing synthetic imagery, significantly improving detail and utility of thermal imaging across various applications. The approach utilizes high-resolution images from the visible spectrum to guide the super-resolution of low-resolution thermal images.
Researchers have developed a system combining bio-inspired cameras with AI to quickly detect obstacles around cars, using less computational power. The hybrid system detects objects up to one hundred times faster than current systems while reducing data transmission and processing needs.
Researchers developed a robot that uses machine learning to automate microinjection in genetic research, enabling large-scale experiments. The technology has the potential to expand genetic research capabilities while reducing costs.
A computer vision researcher has developed privacy software for surveillance videos that obscures identifiable information such as faces and clothing in real-time. The software, funded by the National Science Foundation's Accelerating Research Translation program, aims to balance surveillance needs with privacy concerns.
The DISCount framework combines AI-powered image analysis with human analysis to quickly deliver reliable estimates of building damage and bird flock size. It has been recognized by the Association for the Advancement of Artificial Intelligence for its social impact, winning an award for best paper on AI for social impact.
A team of researchers developed an AI-powered computer vision model to detect Brazilian wild animals on roads and warn drivers in real-time. The system uses roadside cameras and portable computers to identify species such as anteaters, wolves, and tapirs, with the potential to save lives and reduce roadkill.
Researchers from the University of Washington created an AI algorithm to analyze infant poses using limited training data. By leveraging generative AI, they were able to produce high-quality results, enabling parents to monitor their babies' daily activities and detect potential health issues early.
New research combines images with computer-enabled analysis to tackle biological questions globally. Imageomics aims to improve image classification and analysis using machine learning and computer vision, enabling faster scientific discoveries.
Recent deep learning methods have achieved over 97% accuracy in image anomaly detection, but face challenges such as inadequate real-world datasets, inconsistent evaluation metrics, and inefficient loss functions. To improve industrial manufacturing, researchers must address these issues and develop more robust algorithms.
A research group at Chuo University developed a novel non-destructive inspection technique combining multi-functional photo monitoring devices with image data-driven three-dimensional restoration methods. The technique precisely evaluates target objects by compositional identifications and structural reconstructions, providing a breakt...
A new depth from focus/defocus approach, DDFS, combines model-based and learning-based strategies to achieve notable improvements in performance and applicability. The proposed method outperformed state-of-the-art methods in various metrics for several image datasets.
A computer vision system developed by University of Malaga engineers estimates vehicle speeds in real time using a single camera, reducing complexity and costs. The algorithm, published in Neurocomputing, aims to improve vehicle safety and has potential applications in autonomous driving and driver assistance.
Novel Dice loss functions, t-vMF Dice loss and Adaptive t-vMF Dice loss, have been developed to improve image segmentation accuracy in medical images. These new functions outperform conventional formulations and show great potential for critical fields like medical imaging and diagnosis.
A study led by Keck School of Medicine of USC used AI detection technology to analyze influencer content on TikTok between 2019 and 2022, finding an increase in posts that promote e-cigarettes. The prevalence of pod devices, e-juice flavor names, and nicotine warning labels increased significantly over time.
Researchers developed a new 3D inkjet printing system that works with a wider range of materials, including slower-curing materials. The system utilizes computer vision to automatically scan the print surface and adjust the amount of resin deposited in real time.
A recent study by Osaka University's researchers aims to bring science fiction stories closer to reality by studying the mechanical properties of human facial expressions. The team mapped out the intricacies of human facial movements using tracking markers, revealing that even simple motions can be surprisingly complex and nuanced.
Researchers developed 'acoustic touch' smart glasses that translate visual information into distinct sound icons, enhancing the ability of blind or low-vision individuals to navigate their surroundings. The technology significantly improved object recognition and reaching abilities, empowering independence and quality of life.
Researchers at MIT found that similarity-focused generative AI models falter when tasked with designing new products, highlighting the need to prioritize innovation in engineering tasks. By adjusting training objectives and metrics, AI can be an effective 'co-pilot' for engineers, enabling faster creation of innovative products.
Researchers used micro-computed tomography to examine a Rijksmuseum statue and discovered the characteristics of the artist. The study found that the partial fingerprints of the artwork belong to an adult male, corresponding with the attributed model, Laurent Delvaux.
Researchers developed MonoXiver, a new method to help AI extract 3D information from 2D images, making cameras more useful tools for emerging technologies. The method significantly improves accuracy when used in conjunction with existing techniques, such as MonoCon.
Researchers at Osaka University develop a method to train AI models using simulated city images, reducing the need for real data and saving human effort. The approach generates realistic images with accurate ground truth labels, addressing instance segmentation challenges.
A novel AI system developed by City University of Hong Kong improves predictive accuracy in dense traffic, reducing latency and increasing efficiency. QCNet achieves speed and accuracy in predicting road users' movements, even with long-term predictions, making autonomous driving safer and more human-like.
Researchers have developed SMART-BARN, a cutting-edge technology lab for complex behavioral analysis. The facility can host hundreds of animals simultaneously and track their movements in 3D, allowing for unprecedented insights into collective behavior.
Researchers at Pohang University of Science & Technology have developed a sensor technology called computer vision-based optical strain (CVOS) that enhances durability and streamlines fabrication processes. This breakthrough enables the precise recognition of intricate bodily motions through a single sensor.
A team from Nanyang Technological University and the National University of Singapore aims to develop innovative solutions to enhance the accuracy of computer vision systems for autonomous vehicles. They will focus on designing CV systems that can recover to at least 80% of their original accuracy following physical threats or attacks.
Researchers developed a neural network-based system, PAT, for snapshot compressive imaging. It achieves comparable image quality to CASSI and holds strong promise due to advancements in AI processing capabilities.
Researchers developed a pair of modules to enhance polyp segmentation in colonoscopy images, overcoming challenges of image noise and camouflage. The new approach achieved significant improvements in accuracy, with a 2.6% increase in performance and an additional 1.8% gain from a camouflage detection module.
A breakthrough in photonic memory has been achieved, enabling fast volatile modulation and nonvolatile weight storage for rapid training of optical neural networks. The 5-bit photonic memory utilizes a low-loss PCM antimonite to achieve rapid response times and energy-efficient processing.