MIDDLEBURG HEIGHTS, Ohio — Middleburg Heights Police are using facial recognition technology to assist them in shoplifting cases. After one case, on March 30, police received surveillance footage from the BJ’s Wholesale Club on W. 130th Street. Police say the man in a white jacket and red pants took PlayStation headsets out of their boxes, stuffed them into his jacket, and walked out the door with the nearly $500 in loot. Loss Prevention didn’t stop the man. “They didn’t realize it initially, I think it was after the fact,” Middleburg Heights Interim Police Chief Robert Swanson said. An officer had two photos of the suspect’s face from the images provided. “He ran them through a program called Clearview AI that we have,” Swanson said. Clearview AI is a facial recognition software. Swanson says they’ve also used it in felonious assault and human trafficking cases. "It scrapes information from public databases, it goes through millions and millions of images, and it takes it from Instagram, Facebook, a lot of social media platforms, and news articles," Swanson said. But Swanson says his officers only use it as a tool to help identify a person. “We have to per our policy All Clearview AI does is generate an investigative lead. And that kind of points us in the right direction,” Swanson said. The chief says the officers have a mobile version of the technology that lets them take a snapshot of someone while they’re trying to identify them in the field. "It shows that this department is using the tool in kind of a broad range of cases, CSU Law Professor Jonathan Witmer-Rich said. Witmer-Rich says there are real privacy concerns with facial recognition technology, as well as concerns about what happens if it doesn’t identify the right person. “Because you’re looking through
Apr 20, 2026 · via news5cleveland.com
SAN JOSÉ, Calif., April 20, 2026 (GLOBE NEWSWIRE) -- Metafoodx, an AI-powered kitchen intelligence platform for commercial foodservice operations, will showcase its award-winning 3D AI food tracking and analytics technology at the 2026 National Restaurant Association Show in Chicago. The show draws more than 52,000 foodservice professionals and 2,000 exhibiting organizations across 900+ product categories, offering a premier platform for Metafoodx to provide customers and industry leaders with an exclusive look at its breakthrough technology transforming kitchen efficiency and sustainability. Metafoodx is redefining kitchen operations by replacing manual processes and guesswork with real-time, data-driven intelligence. Its proprietary 3D AI scanning system captures critical data points (including precise weight, image recognition, and temperature) in under two seconds per scan. This data is automatically linked to menu items, giving operators a detailed, end-to-end view of food production, consumption, food safety conditions, and waste. The platform integrates seamlessly with leading foodservice systems such as Jamix, Illumia, and Parsley via an open API. At the show, Metafoodx will demonstrate its full kitchen intelligence platform, including the 3D AI scanner and analytics dashboard, which transforms operational data into actionable insights. By analyzing consumption trends and historical patterns, the platform enables kitchens to optimize production, improve forecasting, automate temperature logging for food safety, and significantly reduce overproduction and waste. Customers using Metafoodx have reported up to a 90% reduction in food waste and as much as a 500x return on investment in multiple countries and across C&U, Resorts, Corporate Dining, and QSRs. “The 2026 show is an exciting opportunity for us to connect directly with operators and industry partners, and to demonstrate how our platform brings real intelligence into kitchen operations,” said Fengmin Gong, CEO and Co-Founder of Metafoodx. “We’ve made it simple for teams to use their own data to improve ordering, preparation, and service
Apr 20, 2026 · via markets.businessinsider.com
When police arrested Kimberlee Williams, a grandmother living in Oklahoma, because of a warrant from Maryland, she was shocked. She had never been to Maryland in her life. Ms. Williams later learned that Maryland police had relied on an incorrect result from facial recognition technology that falsely flagged her as a suspect. She is the 14th person in the U.S. to join a growing list of people wrongfully arrested because police let flawed facial recognition technology taint their investigations. Police use of facial recognition technology is dangerous, and stories of people wrongfully arrested because of police reliance on incorrect facial recognition results continue to surface. Today, the ACLU and ACLU of Maryland sent letters to three Maryland police departments on behalf of Ms. Williams, who was wrongfully arrested and jailed for six months because Maryland police relied on a false facial recognition result and concealed their reliance on that unreliable technology from the court when applying for an arrest warrant. One Woman Arrested for a Crime She Didn't Commit On June 23, 2021, Ms. Williams was accompanying one of her daughters on a DoorDash delivery to a local military base in Lawton, Oklahoma. When base security at the entry checkpoint conducted a standard identification check, they discovered outstanding Maryland arrest warrants for Ms. Williams and detained her. These warrants sought Ms. Williams' arrest for a series of fraudulent over-the-counter cash withdrawals in Maryland in December 2019 and January 2020. An unknown individual had entered SunTrust and Truist bank branches in three different counties, impersonated account holders, and fraudulently withdrew thousands of dollars from those individuals' accounts. Ms. Williams, however, was nowhere near Maryland during this time. She was a resident of Oklahoma, living with two of her daughters and their children. While someone was defrauding banks in Maryland, Ms. Williams
Apr 20, 2026 · via creators.com
The Syracuse Common Council tabled a bill on Monday that would prevent business owners from using biometric surveillance systems.
Those systems include facial recognition and eye recognition programs.
Common Councilor Jimmy Monto says lawmakers need more time to look into the bill.
"We're constantly in a place where we're trying to strike a balance between keeping the public safe and also keeping everyone's constitutional rights and their data and their personal information also safe," Monto said. "We deserve both. We should be safe in our homes. We should be safe in streets but also we shouldn't have to walk into a grocery store and worry if someone is scanning our eyes and scanning our face."
Apr 20, 2026 · via spectrumlocalnews.com
Abstract Multi-source information fusion plays a crucial role in enhancing object recognition performance. However, in real-world applications such as autonomous driving and industrial inspection, occlusion and data inconsistencies caused by complex environments can undermine the reliability of extracted features. In this paper, we propose a Topology-Aware Multi-Information Fusion (TMF) model, designed to improve the robustness and generalizability of feature extraction in multi-sensor data. For the first time, our model simultaneously integrates topological architectures into both feature extraction and propagation within a multi-source information framework. The proposed model introduces two key modules. The Enhancing Feature Module (EFM) refines local geometric structures in a topology-preserving manner. The Attention Topology Module (ATM) applies topology-aware attention during feature propagation to dynamically recalibrate feature importance, thereby improving the cross-modal fusion process. In addition, 2D RGB features extracted by a lightweight convolutional encoder are concatenated with 3D point-cloud features, providing a clear and effective fusion strategy. Through a structured fusion framework, our method effectively integrates 3D point cloud features with 2D image-based convolutional descriptors, maximizing the complementary advantages of different sensor modalities. Extensive experiments conducted on the S3DIS and Semantic3D datasets validate the effectiveness of our model. Compared to the typical object recognition model, PointNet, our proposed method achieves a 15.3% and 17.1% improvement in mIoU, respectively. Additionally, we further validate the model using self-collected real-world data, demonstrating its applicability across different data distributions and its potential for real-world multi-modal object recognition. Similar content being viewed by others Data availability This study utilizes both publicly available datasets and self-collected data. The S3DIS dataset is available at http://buildingparser.stanford.edu/dataset.html, and the Semantic3D dataset can be accessed at http://www.semantic3d.net/. The self-collected data were acquired by the authors and are not publicly available. However, they can be provided upon reasonable request to the corresponding author. Code availability The code supporting
Apr 20, 2026 · via nature.com
AI-powered computer vision technology company Restb.ai has announced it now reaches more than 1 million real estate agents through its growing network of MLS partnerships across the United States and Canada. With adoption spanning 26 new MLSs over the past 18 months, Restb.ai’s technology is believed to be one of the most widely deployed AI solutions available to real estate agents in North America. According to a release, Restb.ai’s rapid expansion reflects growing demand from MLSs looking to deliver smarter, more automated tools to their customers. By integrating AI directly into the MLS workflow, real estate agents gain access to a wide range of Restb.ai AI-powered capabilities, including image recognition, automated tagging, compliance insights, and enriched property data, without changing how they work. “These MLSs are leading the way in the deployment of practical and safe AI solutions agents can use right away,” said Dominik Pogorzelski, president, MLS at Restb.ai. “Agents are not being asked to learn new systems. Restb.ai technology is built into the systems they already use every day.” MLSs that have deployed Restb.ai technology for the first time in the U.S. over the past 18 months include: ArkansasONE MLS, Beaches MLS (Florida), Billings Association of REALTORS® (Montana), Charlottesville Area Association of REALTORS® (Virginia), Coeur d’Alene MLS (Idaho), Colorado Real Estate Network, Indiana Regional MLS, Intermountain MLS (Idaho), MLS Technology, Inc. (Tulsa, Oklahoma), MLS United, LLC (Mississippi), Mammoth Lakes Board of REALTORS® (California), MetroList® MLS (California), New Mexico MLS, REALTORS® Association of Indian River County Inc. (Florida), Realcomp (Michigan), Royal Gorge Association of REALTORS® Inc. (Colorado), Greater Alabama MLS Inc, San Francisco Association of REALTORS® (California), Sanibel and Captiva Island Association of REALTORS® (Florida), Tulare County Association of REALTORS ® (California), Western River Valley Board of REALTORS® (Arkansas) and Western Upstate MLS (South Carolina). In Canada, participating organizations
Apr 20, 2026 · via rismedia.com
Many smartphones don’t detect face biometrics spoofs or properly warn consumers Biometric liveness detection remains a significant “flaw” and a “vulnerability” of most Android smartphones with facial unlocking. Most are still prone to simplistic and low-cost spoofs available to inexpert attackers, according to an analysis by Which?. The publication notes that iPhones are generally immune to spoofs with printed 2D photos, due to the depth-sensing capability of Face ID. Some newer Google Pixel devices were also not fooled by flat images in Which? testing. The convenience factor of native device face biometrics is identified as such sometimes, and Which? acknowledges that “some manufacturers have made strides in providing clearer warnings during setup.” Yet many Android smartphones do not, it says, including models from OnePlus and Motorola. OnePlus did just release a new phone with in-display 3D ultrasonic fingerprint biometrics from Qualcomm. Which? labs has tested 208 phones since October of 2022, and found 2D printed photos were good enough spoofs to fool the face biometric unlock systems of 133 devices, or 64 percent of them. Testing during 2025 revealed a 13 percent improvement, year-over-year, after a brutal 2024 in which the share of spoof-prone devices rose dramatically. Samsung’s Galaxy S26 has adequate biometric presentation attack detection (PAD), Which? says, but previous models including the Galaxy S25 do not. At least the manufacturer properly warns consumers that its facial recognition is a convenience feature, rather than a high-security one. While banking apps and digital wallets no longer accept 2D Android face biometrics as a secure authentication factor, Which? warns that users relying on face biometrics to unlock their phone risk a phone thief with their photo reading their text messages, sending emails from their account, which could allow them to reset passwords for other services, access photos and other sensitive documents
Apr 20, 2026 · via biometricupdate.com
Figures Abstract Human action recognition has become increasingly important for applications in security surveillance, healthcare monitoring, and smart environments. However, existing deep learning models typically require substantial computational resources, making deployment on resource-constrained edge devices challenging. To address this limitation, we propose TinyAct, a lightweight framework for real-time human action recognition that combines edge computing with cloud-based processing through knowledge distillation. TinyAct employs a 3D video autoencoder to extract compact spatiotemporal features from video sequences, coupled with classical machine learning classifiers for action prediction. The framework utilizes an AIoT (Artificial Intelligence of Things) architecture where feature extraction occurs on edge devices while classification is performed in the cloud, enabling real-time processing with reduced bandwidth requirements. To enhance performance, we implement knowledge distillation using the ILA-ViT-B/16 transformer as a teacher model to transfer temporal knowledge to our compact student architecture. Our experiments on the Kinetics-400 dataset demonstrate that TinyAct achieves competitive performance while maintaining computational efficiency. Using 16-frame video clips with 1024-dimensional latent features, Random Forest achieved the highest baseline accuracy of 57.00%, followed by SVM (55.00%) and XGBoost (54.00%). The autoencoder-based feature extraction significantly reduces computational overhead compared to end-to-end deep learning approaches while preserving essential spatiotemporal information for accurate action recognition. The knowledge distillation experiments reveal that training configuration critically affects performance, with non-pretrained student models achieving better results (15.11% with SVM) than pretrained ones under teacher supervision. This suggests that joint optimization of the encoder and classifier is essential for effective knowledge transfer in resource-constrained settings. TinyAct’s modular architecture enables flexible deployment across diverse hardware configurations, supporting both lightweight edge inference and cloud-based training pipelines. The framework demonstrates that effective human action recognition can be achieved without computationally intensive deep networks, making it suitable for smart surveillance systems, IoT applications, and embedded devices where computational resources are limited.
Apr 20, 2026 · via journals.plos.org
Przemek Wasinski Has Automated Plane Spotting, via an ADS-B-Tracking Motorized Camera A low-cost SDR dongle picks up planes' transponders, then feeds their location to a motorized camera linked to a Raspberry Pi 4. Developer Przemek Wasinski has made plane spotting an automated affair — by using a low-cost software-defined radio to pick up aircraft transponders and automatically train a camera on their location. "Aircraft in flight continuously broadcast information about themselves, including their location," Wasinski explains of how the project works. "These radio signals use a technology called ADS-B. My system receives these radio signals using a 1090MHz radio antenna. The signals are then processed by a software called Dump1090, which decodes the radio messages and converts the data into JSON format that can be used by plane_tracker.pu." Wasinski build is split into two parts. The first is the tracker itself, which picks up the ADS-B transmissions using a low-cost receive-only RTL-SDR software-defined radio dongle, decodes them, and plots the plane's location on a two-dimensional map laid out like a classic radar scope. The second part then takes this location data and puts it back into the real world — by training a Raspberry Pi HQ Camera Module, connected to a Raspberry Pi 4 Model B single-board computer and with a telephoto lens fitted, on the plane's location. "When an aircraft is close enough Plane Tracker will send its latitude, latitude and altitude to PlaneCam," Wasinski explains. "PlaneCam will then move the camera on the motorized pan-tilt mount to aim at the aircraft and take a picture. The picture taken by the Raspberry Pi Camera will then be analyzed with OpenCV to check whether a plane has been captured, after this the picture and image recognition results will be sent back to PlaneTracker." More information on the project is available
Apr 20, 2026 · via hackster.io
About
Press
Copyright
Contact us
Creators
Advertise
Developers
Terms
Privacy
Policy & Safety
How YouTube works
Test new features
NFL Sunday Ticket
© 2026 Google LLC
Apr 20, 2026 · via youtube.com
DALLAS, April 20, 2026 (GLOBE NEWSWIRE) -- Restb.ai , the real estate industry’s leader in AI-powered computer vision technology, today announced it now reaches more than 1 million real estate agents through its growing network of MLS partnerships across the United States and Canada. With adoption spanning 26 new MLSs over the past 18 months, Restb.ai’s technology is believed to be the most widely deployed AI solution available to real estate agents in North America. Restb.ai’s rapid expansion reflects growing demand from MLSs looking to deliver smarter, more automated tools to their customers. By integrating AI directly into the MLS workflow, real estate agents gain access to a wide range of Restb.ai AI-powered capabilities, including image recognition, automated tagging, compliance insights, and enriched property data, without changing how they work. “These MLSs are leading the way in the deployment of practical and safe AI solutions agents can use right away,” said Dominik Pogorzelski , President, MLS at Restb.ai. “Agents are not being asked to learn new systems. Restb.ai technology is built into the systems they already use every day.” MLSs that have deployed Restb.ai technology for the first time in the U.S. over the past 18 months include ArkansasONE MLS , Beaches MLS (Florida), Billings Association of REALTORS ® (Montana), Charlottesville Area Association of REALTORS ® (Virginia), Coeur d’Alene MLS (Idaho), Colorado Real Estate Network , Indiana Regional MLS , Intermountain MLS (Idaho), MLS Technology, Inc. (Tulsa, Oklahoma), MLS United, LLC (Mississippi), Mammoth Lakes Board of REALTORS ® (California), MetroList ® MLS (California), New Mexico MLS , REALTORS ® Association of Indian River County Inc. (Florida), Realcomp (Michigan), Royal Gorge Association of REALTORS ® Inc. (Colorado), Greater Alabama MLS Inc , San Francisco Association of REALTORS ® (California), Sanibel and Captiva Island Association of REALTORS ® (Florida), Tulare County Association
Apr 20, 2026 · via markets.businessinsider.com
A new 'Perspective' article says generative AI may help scientists read cancer’s hidden complexity across images, molecules, and clinical data, opening a possible new path to smarter diagnosis, discovery, and treatment. Perspective: Tackling the complexity of cancer with generative models. Image Credit: Antonio Marca / Shutterstock A recent Perspective article published in the journal Cell argues that generative models could help address the complexity of cancer. The “Hallmarks of Cancer” provided a framework to systemize the understanding of cancer biology. They proposed a set of principles dictating the transformation of normal cells into malignant cells and subsequent cancer progression. The hallmarks represent a reductionist framework that has unified diverse observations, yielding valuable insights. However, an intentionally simple framework cannot adequately explain the multifaceted mechanisms of cancer. Thus, complementary tools are required to capture the complex, multiscale, and multimodal nature of cancer. In this paper, the authors proposed that generative models built on advances in artificial intelligence (AI) can address the complexity of cancer. AI for Cancer Detection and Biological Understanding AI has achieved significant strides in its ability to model complex patterns over the years. Advances in learning algorithms, data availability, and processing power have led to human-level or even higher accuracy in some tasks. The applications of AI to cancer include understanding, detection, and intervention. Much of the progress in AI for cancer has been in detection. The development of deep convolutional neural networks has significantly improved image classification performance. Examples include breast cancer detection using mammographic data, skin cancer classification using lesion images, and lung cancer detection using computed tomography data. Further, many advances in understanding cancer biology have resulted from improvements in its molecular characterization. As the value of epigenomics, proteomics, transcriptomics, and other -omics measures has become clear, there is growing interest in characterizing their high-dimensional
Apr 19, 2026 · via news-medical.net
Connections Unlimited Challenges Players To Group Words By Hidden Links Ellen Smith — April 19, 2026 — Tech References: connectionsunlimitedonline.github.io Connections Unlimited is a word-based puzzle game focused on identifying relationships between groups of terms. Players are presented with sets of words and tasked with organizing them into categories based on shared characteristics or hidden connections. The gameplay emphasizes vocabulary, pattern recognition, and associative thinking, often drawing on general knowledge and cultural references. It is designed as an ongoing experience with no fixed endpoint, offering continuous challenges rather than level-based progression. The game is typically used for casual entertainment as well as cognitive engagement, appealing to users interested in language and logic puzzles. It reflects a broader category of word games that prioritize mental agility and replayability. By removing ads and usage limits, the platform focuses on uninterrupted interaction. Its primary function is to provide an accessible and repeatable puzzle format centered on discovering conceptual links between words. Image Credit: Connections Unlimited The gameplay emphasizes vocabulary, pattern recognition, and associative thinking, often drawing on general knowledge and cultural references. It is designed as an ongoing experience with no fixed endpoint, offering continuous challenges rather than level-based progression. The game is typically used for casual entertainment as well as cognitive engagement, appealing to users interested in language and logic puzzles. It reflects a broader category of word games that prioritize mental agility and replayability. By removing ads and usage limits, the platform focuses on uninterrupted interaction. Its primary function is to provide an accessible and repeatable puzzle format centered on discovering conceptual links between words. Image Credit: Connections Unlimited Trend Themes 1. Continuous Play Cognitive Puzzles - Endless, non-level-based puzzle loops create new formats for sustained user engagement that can redefine retention metrics for casual games. 2. Ad-free Microlearning Entertainment -
Apr 19, 2026 · via trendhunter.com
German Society for Internal Medicine: "Treating people, not data" Staff shortages and more endanger quality of care, according to internists' congress. AI is supposed to fix it, but humans must always remain the focus. At the 132nd Congress of the German Society for Internal Medicine, physicians discussed the use of artificial intelligence, which goes far beyond pure pattern recognition. AI agents are intended to actively relieve doctors and nursing staff of routine tasks, thus counteracting staff shortages and increasing bureaucracy. At the same time, modern technology enables more differentiated diagnostics, the experts emphasized. Artificial intelligence in everyday hospital life is evolving from a consulting to an actively acting system. “We are moving from 'not just advice, but also action',” explained Prof. Jens Kleesiek, Director of the Institute for Artificial Intelligence in Medicine at Essen University Hospital, at a press conference for the congress. Given staff shortages and administrative overload, the use of such technologies is unavoidable: “We can no longer guarantee the quality of care, the security of supply, if we do not use further tools.” Videos by heise As a concrete example, Kleesiek cited “agentic AI,” which autonomously coordinates complex processes such as patient discharge or warns staff if an allergy has not been correctly noted in the record. At the same time, he cautioned that with all the technological support, “common sense should not be switched off.” The danger is to blindly trust technology, similar to a navigation system that drivers follow and “drive into the river or into a field somewhere.” The guiding principle must therefore be clear, according to Kleesiek: “We treat patients, not data.” Re-evaluation through data: The BMI under scrutiny The advancing digitalization also leads to a re-evaluation of established medical parameters. The complexity of metabolic research reaches far back into evolutionary history, as
Apr 19, 2026 · via heise.de
About
Press
Copyright
Contact us
Creators
Advertise
Developers
Terms
Privacy
Policy & Safety
How YouTube works
Test new features
NFL Sunday Ticket
© 2026 Google LLC
Apr 19, 2026 · via youtube.com
The Internet remembers more than most people realize. Every photo that appears online can remain available for years, often without the person's knowledge. PimEyes, one of the most advanced face search tools on the Internet, responds to this reality. At first glance, it seems like a useful tool. In fact, however, it raises very sensitive questions about privacy, anonymity and control over one's own digital identity. The principle of operation is very simple. The user uploads a photo of a face and the system then uses artificial intelligence to search the publicly available Internet. Within a few seconds, it can display places where the same or very similar face can be found. Unlike conventional image searches, the technology does not focus on the context of the photo, but purely on biometric facial features. The result is links to specific websites where the person appears. The service itself claims that it does not identify specific people as such, but only connects similar photos across the Internet. According to the authors, PimEyes is primarily intended to have positive uses. Users can check where their photos are located and, if necessary, resolve their removal. At a time when content spreads extremely quickly and often without consent, it is a tool that can help protect one's own identity and digital footprint. At the same time, however, it is also necessary to speak openly to the other side. The same technology can be very easily abused. All it takes is a photo of a stranger and within a moment it is possible to find other images, social profiles or the context in which the person appears on the Internet. This significantly increases the risk of abuse, whether it is stalking, invasion of privacy or revealing identity without consent. This accessibility is a fundamental difference compared
Apr 19, 2026 · via letemsvetemapplem.eu
BlabbyAI Speech to Text is a voice input tool designed to convert spoken language into written text across desktop and browser environments. It enables users to dictate content directly into websites or applications, supporting workflows such as email writing, note-taking, and messaging. The platform offers multilingual transcription capabilities, accommodating a wide range of global users and use cases. By replacing manual typing with voice input, it aims to improve speed and accessibility in content creation tasks. It is typically used by professionals, content creators, and individuals seeking more efficient input methods or hands-free operation. The tool reflects a broader trend in productivity software toward speech-based interfaces, driven by advances in automatic speech recognition technology. Its primary function is to streamline text entry processes and reduce reliance on traditional keyboard-based input across digital platforms. Speech Recognition Tools BlabbyAI Enables Voice Typing Across Websites And Applications Trend Themes 1. Voice-first Productivity - Real-time dictation replacing keyboard input in everyday workflows creates potential for interfaces that prioritize speech as the primary mode of content creation. 2. Multilingual Speech Interfaces - Support for many languages opens possibilities for tools that normalize cross-language collaboration through accurate, localized transcription and translation layers. 3. Cross-platform Voice Integration - Seamless voice input across browsers and desktop applications suggests ecosystems where voice capabilities are universally available regardless of platform boundaries. Industry Implications 1. Enterprise Software - Embedding speech-to-text into productivity suites could substantially reduce meeting note burdens and accelerate internal documentation lifecycles. 2. Education Technology - Classroom and remote-learning platforms equipped with reliable transcription can expand accessibility and create richer study materials for diverse learners. 3. Customer Service & Contact Centers - High-quality, real-time transcription in support systems enables faster resolution metrics and more accurate capture of customer interactions for analytics.
Apr 19, 2026 · via trendhunter.com
Testing has revealed that smartphones from major brands including Samsung and Motorola can be unlocked using printed images of the owners face. Consumer experts have warned that this security loophole could be exploited by criminals to unlock devices and access personal information. Which?, the UK’s largest independent consumer organisation, has been carrying out lab tests on hundreds of phone models since 2022, assessing features from battery life to security features. While most smartphones now offer facial recognition to unlock mobile devices, Which?’s testing found that as many as 64% of phones – 133 devices since 2022 – could be tricked by a printed image. Phones that failed the organisation’s tests include top-of-the-range models such as the Oppo Find X9 Pro, which retails for upwards of £1,000. According to Which?, Samsung’s former flagship range of Galaxy S25 phones could all be fooled by a 2D photograph, while the some lower-priced Android phones also struggled, often relying on a standard 2D facial recognition system. Which? Found that the newest Google Pixel models and Samsung Galaxy S26 series passed their tests and Apple Face ID was considerably harder to trick. The non-profit says that some device cameras take flat pictures, which means they cannot always distinguish between a real person and a high-resolution photo. Most phones that failed Which?’s test did have an onscreen message warning that face recognition cannot be relied upon for security. Lisa Barber, Which? Tech Editor said: “These security flaws are far from isolated incidents. “The majority of Android phones we’ve tested in the last four years can be easily unlocked using a 2D image, and some manufacturers are still failing to adequately warn their users that this is the case. “We’d urge affected users to set up alternative methods of security, like a fingerprint or a PIN, which
Apr 19, 2026 · via news.stv.tv
Specialists from the Unified Center for Biometric Testing (ECBI) conducted the first testing of Russian facial recognition systems. Seven IT solutions were tested for accuracy, the press service of the Department of Information Technology (DIT) of Moscow reported. The tests involved four companies from Moscow and one from Chelyabinsk. Their biometric algorithms had to find similar faces in a database of 750 thousand images. How the algorithms were evaluated When evaluating IT solutions, the following parameters were taken into account: data collection reliability (processing images of a person in glasses, in a half-turn) and recognition accuracy (error rate). The final result was calculated based on the arithmetic mean of all metrics. Dmitry Golovin, Deputy Head of the DIT of Moscow, said that the algorithms analyzed images from city video surveillance cameras. The test participants demonstrated their developments in real, not laboratory conditions, he added. As a result, the market receives a transparent tool for selecting technologies for a specific business task. The DIT of Moscow also noted that the ECBI will regularly conduct such tests. This will allow tracking progress in the development of domestic biometrics. Read more materials on the topic: - In Russia, they began to let into trains using biometrics — how it works - Face instead of a passport: biometrics is gaining popularity in Russia - Payment by "face" in buses: a new biometric travel system will be launched in Moscow
Apr 19, 2026 · via www1.ru
Abstract At present, multi-scale learning is a popular and effective method to improve the accuracy of real-time semantic segmentation. However, these multi-scale methods do not consider the influence of CNN receptive field on the feature discriminability. They still suffer from the small receptive field and insufficient feature extraction, which limits the accuracy of real-time segmentation. To solve the problem, we propose a novel Triple-Branch Multi-Scale Network (TBMSNet) for real-time semantic segmentation. Specifically, in initial feature extraction stage, we propose the Simple Inverted Residual (SIR) module with reducing the number of input channels, using two successive SIR modules to initially extract features, which can enhance the feature extraction capability of lightweight backbone network. Subsequently, we design a new multi-scale triple-branch structure to parse detail, semantic, and boundary information respectively. In triple-branch structure stage, we propose the Dilation-wise Residual (DWR) module in Semantic branch, which combines the multi-scale detail and boundary branches down-sample semantic feature maps to 1/64. The design can extend the valid receptive field and improving the capture efficiency of multi-scale information. Besides, we also design a Multi-scale Semantic Aggregation Pyramid Pooling (MSAPP) module in semantic branch, which connects multi-scale pooling maps before the convolutional layer to form local and global context representations for further enriching the semantic information, and extract multi-scale features more efficiently. Experiments show that our TBMSNet achieves an accuracy of 80.5% mIoU on Cityscapes and an inference speed of 50.4 FPS, achieving the best trade-off between inference speed and accuracy. Similar content being viewed by others Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Corresponding authorshould be contacted if someone wants to request the data from this study. Code availability The code are available from the corresponding author upon reasonable request. References Long et
Apr 19, 2026 · via nature.com