Dive Brief: - Skilled labor demand has almost doubled over the past three years, driven in part by a rise attributed to AI, according to results of a National Fire Protection Association survey released Tuesday. The organization sets standards for fire and electrical safety in the built environment. - The survey suggests a rise in demand for skilled labor to work on data centers and other types of AI-related infrastructure, pulling trained professionals away from other settings. Roughly a third of respondents to NFPA’s survey cited increased demand for skilled labor related to AI infrastructure, according to NFPA’s survey of 326 skilled trades workers and others in the field, including facility managers. - The results cause concern that as AI infrastructure is prioritized, “skilled workers begin leaving other jobsites to satisfy those needs,” Kyle Spencer, director of NFPA LiNK, said in a statement. At the same time, AI could help workers adapt to these rapid shifts and fill those gaps, with workers already using AI to quickly ensure their work meets the latest code requirements, NFPA says. Dive Insight: The rapid scaling of data centers and digital infrastructure in the U.S. is exacerbating challenges in hiring skilled labor across industries, according to a March report by Randstad North America, which analyzed more than 150 million job postings for key roles between 2022 and 2026 to identify labor demand. For example, it’s now more time-consuming to hire an HVAC professional or an electrician than a software developer. Contributing to that shortage is a generational shift, with seasoned workers retiring and taking the skills that have helped prevent downtime and costly errors with them, Randstad says. Despite a looming retirement cliff for the engineering and technical workforce, vocational training pipelines haven’t kept pace with demand. Conservative estimates suggest a 30% skills gap
Aug 19, 2026 · via facilitiesdive.com
DHS tries to lessen constraints of biometric identity verification The Department of Homeland Security is working to mitigate technical challenges tied to biometric identity verification systems, per documents posted Tuesday and earlier this month. DHS built the Biometric Data Collection and Verification System to quantify the performance of data acquired and stored by the agency. The system can identify deficiencies and provides standardized testing measures. In addition to tackling the challenge of identifying failures or inaccuracies in biometric matching algorithms, the BDCVS is meant to speed up data acquisition and delayed retrieval speeds, according to the agency. DHS characterized BDCVS as a “proven system” that improves trust in biometric matching outcomes with applications across screening venues. The agency also developed another new system that aims to address additional shortcomings of typical biometric identification tools used for traveler screening. Biometric verification systems can struggle to properly identify individuals with common first and last names or when birthdays are shared by multiple people, DHS said in its research note. Data input errors, such as misspellings or use of nicknames can also present roadblocks, leading to delays in verification, significant bottlenecks for security personnel and travelers and extended screening process time. In response, researchers from the Transportation Security Administration and the DHS’s Science and Technology Directorate created the Biometric Identity Disambiguation system. BID determines whether an individual matches their identification information by comparing the person’s biometrics and other information, like a driver’s license, to what’s stored in accessible databases. The system then assigns a score that indicates how close of a match the individual is to the documents provided, signalling what level of additional checks are needed to screening personnel. DHS said the technology has potential as a “proven system” across border checkpoints, testing and registration centers and as part of transportation, event
Aug 19, 2026 · via fedscoop.com
HOA surveillance cameras secretly installed for ‘neighborhood security’ draw massive backlash — and are killing home sales See more of our coverage in your search results. Add The New York Post on Google Big brother is watching your open house. Homeowners associations across the country are quietly signing up for Flock Safety’s license plate cameras, and the backlash is now strong enough to blow up home sales, kill signed contracts and land cities in court, with New York at the center of the fight. Nowhere is the backlash more raw than in the Adirondacks, where the fight has already toppled a signed contract and sparked open revolt inside a gated community. In Saranac Lake, a resident living behind a gate who protested the cameras in his own neighborhood said the issue isn’t security itself. It’s who gets to look. “The problem I have is not with security cameras. It’s with the Flock system directly. Anyone and their mother can have easy access to this database. The way the data is readily being shared. It is completely unconstitutional,” he told The Post. That anger boiled over at a packed Saranac Lake village board meeting earlier in the year after officials quietly signed a contract for a dozen license plate readers and security cameras using a state grant. Residents said they’d learned about the rollout only after cameras started going up on telephone poles around town. “These cameras are not going to keep us safe,” resident Sandra Kalinowski told the board, drawing applause from the crowd. Another resident, David Lynch, pushed the board to reverse course entirely. “First I believe these cameras are currently in violation of Saranac Lake Police Department policy. Secondly, I believe the public has been misled about Flock’s access to our data. My ask of you tonight is
Aug 19, 2026 · via nypost.com
NEWS
Daytona Beach debates police use of driver’s license photos in facial recognition cases
Daytona Beach City Commissioners will decide Wednesday whether police can continue using a statewide database of driver’s license photographs as part of facial recognition investigations.
Aug 19, 2026 · via clickorlando.com
DAYTONA BEACH, Fla. – Daytona Beach City Commissioners will decide Wednesday whether police can continue using a statewide database of driver’s license photographs as part of facial recognition investigations. The vote, scheduled for 6 p.m., comes as law enforcement agencies increasingly use technology to help identify people suspected of crimes — while privacy advocates and residents question how far that technology should go. In Daytona Beach, police use facial recognition software to compare an image of a person to photographs in large databases, including driver’s license photos. The technology itself does not make an arrest or determine that someone is a suspect. Instead, the software generates potential matches that must then be reviewed by a person. For some Daytona Beach residents and visitors, facial recognition is simply another investigative tool. Embry-Riddle student JP Nass said he sees a practical benefit to the technology. “It’s easier to track and see who’s who. It’s better to catch people doing bad things.” Nass said he believes facial recognition should have a legitimate investigative reason behind its use rather than being used simply to identify people. “Only if you have a reason to do it.” But other residents are more concerned about the amount of information law enforcement can collect and how that information is used. Daytona Beach police also use Flock cameras, which can collect information about vehicles traveling through the city. Brian Turnipsed said he is uncomfortable with that type of surveillance, particularly when the technology is operated by a third party. “That I feel is an invasion of privacy. There’s no one manning them. They’re just out there run by a third party.” Turnipsed said there should be limits on how law enforcement tracks people, particularly when there is no probable cause. “I still feel like if they’re going after you,
Aug 19, 2026 · via clickorlando.com
Facial Recognition Startup ClarityCheck Left 9 Million Face Photos on an Unsecured Server A people-search tool that promises to identify anyone from a single photograph left more than 9 million image files, including close-ups of adults, teenagers, and children, sitting on an unsecured cloud server for months, according to research published this week. The database, traced to the facial-recognition service ClarityCheck, contained roughly 450 GB of images stored in an Amazon S3 bucket without password protection or encryption. Anyone with the right URL could have browsed folders labeled "faces" and "profiles," downloading what appeared to be profile pictures, screenshots, and other photographs. A second misconfiguration also exposed email addresses and phone numbers through the company's website. The findings come from Jeremiah Fowler, an independent security researcher who has documented a long string of exposed databases in recent years. Fowler told WIRED that he discovered the open bucket while investigating ClarityCheck and that his initial attempts to notify the company went unanswered. The database was only secured after WIRED contacted ClarityCheck in July. ClarityCheck is among the growing number of so-called people-finder tools that have proliferated online. The company's website advertises searches by phone number, email address, vehicle identification number, and name. Its photo-search page claims it can "identify anyone in a photo" and locate social media profiles "in seconds." A WIRED reporter who tested the service with their own face watched as the site said it was "scanning facial landmarks" and "mapping unique face geometry" before returning a report that included the reporter's full name, biography, and links to multiple online photos. The company offers deeper reports, including "hidden dating profiles," for a fee. The site requires users to attest that they have permission to upload any photo they submit. But Fowler argues that this safeguard is essentially meaningless for
Aug 19, 2026 · via finance.biggo.com
The story so far: The Supreme Court on Tuesday (August 18, 2026) said it will examine the “proportionality” of the use of facial recognition system (FRS) deployed during the students’ protest in the national capital last month after the Delhi Police in an affidavit admitted to its use. The police said that “FRS does not automatically capture, create, generate, or maintain profiles of every individual present at the protest site, nor is it deployed for indiscriminate surveillance or collection of personal information of peaceful protesters unless he has a previous criminal record,” and assured that the use of the technology was “legitimate, bona fide, and proportionate policing measure”. However, the very nature of the system and its possibilities that, if left unchecked, could have extreme implications for public behaviour in the country is concerning. How does facial recognition work? While the use of the AI-powered product of machine learning to monitor a mass gathering is something new for many, facial recognition has been part of the everyday life of human beings around the world for some time. From screen locks in mobile phones to contactless air travel with DIGI Yatra initative, Indian are getting increasingly exposed to various degrees of FRS. The use of the technology can be broadly classified into two avenues, verification and identification. The most common use that has been incorporated into daily lives is verification where the system confirms the identity of a person from previously provided datasets, like in the case of mobile phone security systems. In case of identification processes, the system scans the distinguishable features of people and runs it against a database to determine their identity without them having to reveal it. Verification using FRS is voluntary while identification may or may not be done with the consent of those under the
Aug 19, 2026 · via thehindu.com
Bloomington settles lawsuit with man wrongfully arrested due to facial recognition technology The City of Bloomington, Minnesota, has settled a lawsuit brought by a man who said he was wrongly arrested back in 2021. Kylese Perryman claims his rights were violated, saying investigators carelessly and incorrectly identified him as the man responsible for a violent felony. He spent five days in jail, 30 days on home monitoring and, after 52 days, charges were eventually dropped His lawyers say this is a case of an investigator comparing surveillance photos of a suspect in a robbery and carjacking to a booking photo of another man and ultimately arresting the wrong person, ultimately the result of faulty facial recognition technology. "So, it is not a positive match, and that is all it does," said Dr. Manjeet Rege, director of the Center for Applied Artificial Intelligence at the University of St. Thomas. "It generates a lead." Lawyers say that Perryman was not near the crime scene. A home surveillance image shows Perryman at a party while two other men carjacked someone, robbed some women and then used those stolen cards at Walmart. Surveillance from Walmart showed the suspect does not have tattoos. Perryman does have tattoos. "I think it becomes a great benefit if it is utilized properly in sync with other regular law enforcement tools," Rege said. "It may become a risk if there are flaws in the process in which it is implemented." Law enforcement's failures alleged in the suit include its failure to conduct a lineup or to contact eyewitnesses to confirm the identification, to consider Perryman's verifiable alibis, to note distinct physical differences between Perryman and the suspect and to promptly drop charges against Perryman when his defense attorney produced clear evidence of his innocence. "I brought this lawsuit because
Aug 19, 2026 · via cbsnews.com
Amnesty International slams rise of Argentina’s 'surveillance state' Massive online surveillance is deterring the right to protest in Argentina, says Amnesty International NGO in latest report. Amnesty International Argentina on Tuesday denounced the growing use of surveillance technologies by President Javier Milei's government against his critics. In a new report (entitled Estado de vigilancia: la expansión del uso de tecnologías de vigilancia masiva por el gobierno de Argentina, or "Sensing the Surveillance State: Argentina’s Expanding Policies and Technologies of Mass Surveillance") Amnesty said the creation of databases of protesters, the monitoring of their social media and the use of facial recognition technology was stifling dissent. Between December 2023, when Milei took office, and December 2025, the Argentine government "implemented a slew of security and intelligence reforms and expanded their use of surveillance technologies," the report said. "These measures constitute an intensification of state monitoring capacities... aimed at controlling and deterring protest and increasing authoritarian practices," it added. Milei's drastic austerity measures have repeatedly sparked mass protests in Buenos Aires and other cities. Amnesty said that the National Security Ministry spent at least US$1.2 million to acquire surveillance technologies between 2024 and 2025. In interviews with Amnesty, a group of journalists, activists, rights defenders and some of the retirees who demonstrate weekly for higher pensions and better healthcare access all said they felt they were being watched. Surveillance technologies “have become an integral part of the state’s control infrastructure,” said Matt Mahmoudi, an Amnesty International researcher and adviser on artificial intelligence and human rights. The Amnesty report said the government had purchased a licence to use facial-recognition technology from Clearview AI, a company “known for the mass exploitation of biometric data without a sufficient legal basis.” In recent years, Clearview has been fined by authorities in France, Greece, Italy and the
Aug 19, 2026 · via batimes.com.ar
Explore the things
you love
.
Log into Facebook
Email or mobile number
Password
Log in
Forgot password?
Create new account
English (US)
Español
Français (France)
中文(简体)
العربية
Português (Brasil)
Italiano
More languages…
Sign Up
Log In
Messenger
Facebook Lite
Video
Meta Pay
Meta Store
Meta Quest
Ray-Ban Meta
Meta AI
Instagram
Threads
Privacy Policy
Consumer Health Privacy
Privacy Center
About
Create ad
Create Page
Developers
Careers
Cookies
Ad choices
Terms
Help
Contact Uploading & Non-Users
Meta © 2026
Aug 19, 2026 · via facebook.com
Eyedaptic recently announced results from an independent real-world clinical evaluation demonstrating substantial immediate improvements in functional vision among patients with advanced retinal disease. The findings were presented by David Almeida, MD, MBA, PhD, FRCSC, at the American Retina Forum Annual Meeting, which took place Aug. 5-8 in South Lake Tahoe, California. Let’s start with Eyedaptic. Specializing in vision-enhancement technology, the privately-held company’s products are designed with wearable vision aid software enabled by AI and augmented reality (AR) hardware. Unlike therapies created to slow disease progression, Eyedaptic’s technology is intended to enhance functional vision by digitally optimizing visual information in real time, helping patients utilize their remaining peripheral vision more effectively. Explain how this is done. Through the proprietary Simulated Natural Vision Software, which works by optimizing a patient’s functional (remaining) peripheral vision via pixel manipulation—“simulating” clearer vision. - The intent: To give patients an improved quality of life with a hands-free, easy-to-wear glasses design (take a look at the clinical research supporting the tech). Which patients is this technology designed for? The company is currently targeting patients diagnosed with age-related macular degeneration (AMD) and diabetic retinopathy (DR). And that software is included in these smart glasses? Indeed. In fact, it’s included in Eyedaptic's entire line of AI-enhanced eyewear: the EYE5, EYE6—and as of March 2026—the EYE7. Among the features of these glasses: - Lightweight design (weighing in at less than 3 oz) - Facial detection capabilities - Auto-zoom mode for viewing text in various situations - Contrast enhancement controls (for additional image enhancements) - 2-in-1 wearable and hand-held magnifier - EyeSwitch technology As a side note: Prior iterations of the company’s products also encompassing this patented technology include the EYE2, EYE3, and EYE4. Are they cleared by the FDA? According to Eyedaptic, its smart glasses are Class I exempt medical
Aug 18, 2026 · via glance.eyesoneyecare.com
Pulp nonfiction A 404 investigation reveals that Amazon is buying used books in bulk and scanning them in a warehouse in Las Vegas—often destroying the texts in the process. • 3 min read TL;DR: For months, booksellers wondered why they were receiving bulk orders for a bunch of obscure used books. Now they may have an answer: An Amazon facility in Las Vegas appears to be processing used books and digitizing them, presumably to train its AI. What happened: 404 Media says it got a bookseller to plant an Apple AirTag in a book that was a part of a large order. The AirTag took a rather circuitous journey, from California to Milwaukee, then Colorado, before it arrived at an Amazon facility in Vegas. That facility contains a previously unknown warehouse called VGT3, marked by a logo of a menacing dinosaur gripping a book. There, workers spend their time scanning rare books—sometimes destroying them in the process. We don’t know much else about the operation: An Amazon spokesperson only confirmed to 404 that the company “purchases books through commercial channels to help develop and improve the products and services our customers use.” The “rare” used books mystery: Booksellers noticed the unprecedented uptick in used book sales because of the order sizes and the kind of books being purchased. The texts are considered “rare” because they cover niche topics and few copies remain—an example being How to Use Corel WordPerfect 1991, per one bookseller. So we’re not talking about a first-edition James Joyce. Still, the trend couldn’t definitively be connected to AI firms until now. Tech Brew breaks down the biggest tech news, emerging innovations, workplace tools, and cultural trends so you can understand what's new and why it matters. By subscribing, you accept our Terms & Privacy Policy. Not just
Aug 18, 2026 · via techbrew.com
By Lauren Wilkin Meta has filed a patent for new AI glasses that use facial recognition to capture moments without a command from the user - building up a searchable memory bank. The patent, filed by Meta on August 13, covers the invention of an AI assistant built into a pair of smart glasses that can decide what to capture, when and what to store as a 'memory' - all without the wearer's input. This means the glasses could record events without the wearer commanding them to, focusing on 'interesting' events based on facial expressions, conversations, behaviors and where the user is looking. The AI assistant would also build up a profile of relationships, based on facial recognition - pulling up certain records associated with each person the wearer interacts with. The patent says that the glasses could then "access a plurality of episodic memories associated with the user." In a real-world example, the glasses could capture moments throughout a dinner party, focusing on and zooming in on interesting conversations and events. At the end of the night, it could deliver a 'highlight' file to the wearer, containing videos and pictures captured throughout the evening. The wearer would also be able to ask the assistant "what happened at this time in the dinner party?" and the AI assistant would be able to play back the video footage captured. The patent says that determining what to capture is based on a measure of "interestingness" which can be influenced by what someone is doing, their facial expression, objects they're associated with, and even where the wearer is looking using "eye gaze data." This would mean wearers would have a searchable memory bank, where they are able to recall conversations and events through video and photos captured by the smart glasses. Importantly, this
Aug 18, 2026 · via nbcrightnow.com
Londoners are concerned about the rollout of facial recognition technology (FRT) due to worries about errors, privacy and surveillance, recent polling has revealed. Meanwhile the British Transport Police (BTP) has admitted its trial at railway stations, which has been running for six months, has caught no serious offenders as a result of using the technology, though they say the threshold has been set deliberately high in terms of the offenders it’s being used to find. 72 per cent of respondents to the polling in Greater London want the public to have a say on how FRT is used, while two-thirds worry errors could get people into trouble unfairly. Another 62 per cent are concerned about how images of faces will be stored, while half feel there is a risk of racial bias. The polling, carried out by Opinium earlier this year, comes amid the start of live facial recognition (LFR) cameras being rolled out at some London Tube stations. It was first deployed at Victoria Underground station as part of a trial that started in February, and will rotate between different stations until November. While the majority of Londoners – 70 per cent – support the use of FRT by police forces, more than half say it should only be deployed where there are a clear public benefit and strong safeguards in place. Tony Kounnis, CEO of Face Int, which commissioned the research, said: “It’s far too simplistic to say that people in London are for or against facial recognition. This research shows that it’s far more nuanced than that – people have specific and understandable concerns, particularly the accuracy of the technology, their own privacy, and the security of biometric data. “Public trust cannot be overlooked; nor is the efficacy of the technology enough. Organisations deploying FRT need to
Aug 18, 2026 · via standard.co.uk
The Delhi Police submitted this in an affidavit sworn by DCP Sachin Sharma, which stated the crowd turned violent and breached many layers of barricades in bid to march towards Parliament. On facial recognition software, Delhi Police said that the use of the technology was a proportionate policing measure. The affidavit said the software does not automatically capture the profiles of every individual present at the protest site, nor is it deployed for indiscriminate surveillance or collection of personal information of peaceful protesters unless they have a previous criminal record. The affidavit has been filed in response to petitions alleging police excesses against Cockroach Janta Party and other protesters at Jantar Mantar. Solicitor general Tushar Mehta told a three-member bench headed by CJI Surya Kant that the technology captures the faces of only those persons whose details are recorded in the crime database. The affidavit further claims that no action is taken solely on the basis of facial recognition software. Field verification is also carried out to ascertain whether the person in question was present at the site. Detailing the sequence of events, the affidavit said although permission for a dharna was granted only for June 20, post 5 pm, the protesters did not vacate the venue. The affidavit added the agitation escalated after a call was given for a march towards Parliament on July 20. It further said "permission was neither sought nor granted for any 'Chalo Sansad' march" and that prohibitory orders under Section 163 of BNSS had been in force in New Delhi since June 21.
Aug 18, 2026 · via economictimes.indiatimes.com
The recent use of artificial intelligence-equipped surveillance cameras by Montreal police is raising questions about privacy rights and how the technology will be used, as the use of similar cameras in the United States has become a major issue. The Genetec Cloudrunner cameras are similar to those made by company Flock Safety, an AI surveillance system widely used in the U.S. that has faced widespread public resistance. But how do the two types of cameras differ, and what information are the smart cameras providing to the Montreal police? The Gazette spoke with experts in AI ethics and a representative from Genetec to answer to growing concerns over automatic licence plate reading cameras. What, exactly, makes these cameras controversial? While traditional security cameras only create video, automated licence plate reading (ALPR) cameras can capture information about the type of car, colour of car, direction of travel, and the licence plate, explained Andrew Elvish, an executive at the Montreal-based company Genetec. “It’s looking at the outside of the vehicle. So it’s designed specifically to do that,” he said in an interview. In the U.S., the Flock cameras can tie the licence plate information to a person’s identity, which can be searched in a centralized database system shared between the camera operators. But according to Elvish, Genetec doesn’t have a central database shared with all the company’s camera users in the same way Flock does in the U.S. “We do not as a company sell access to data. Our business model isn’t predicated on aggregating and then selling data to customers, or giving customers as a benefit access to a vast amount of data,” Elvish said. It’s that shared database that’s caused so much concern in the U.S. “People don’t like surveillance. But when you tie surveillance to location, and you tie surveillance
Aug 18, 2026 · via montrealgazette.com
Figures Abstract Protecting facial recognition privacy is crucial amid deep fake threats, biometric risks, and third-party database access concerns. Despite many recent face recognition methods achieving high accuracy, most existing works either ignore privacy protection or apply privacy mechanisms without designing CNN structures that effectively learn from heavily perturbed facial data. This research introduces a secure face recognition system based on Differential Privacy (DP), employing a Convolutional Neural Network (CNN) and face classifiers. In this study, we develop a CNN through the incorporation of multiple batch normalization layers. This CNN is capable of recognizing the randomized image of the DP technique. To ensure privacy, the face database undergoes perturbation using DP techniques before releasing to any unauthorized access. The CNN model learns from these perturbed images, extracting features that are subsequently used by a predictor to classify the face. The CNN model learns from these images and then this trained CNN extracts features from an image that needs to be recognized. Ultimately, a predictor classifies this face. We evaluate three DP techniques namely Laplacian, Gaussian, and DP-blur using four predictors to evaluate the privacy-preserving capabilities of the proposed method. Each DP technique is evaluated by varying privacy parameters from 0.5 to 8 with an interval of 0.5. This research employs two datasets, namely LFW and IC. The DP blur with Logistic regression predictors provides the highest privacy, achieving excellent accuracy rates of 97% and 77% for these datasets. This outcome surpasses all baseline methods. The research offers an in-depth analysis of various DP techniques to construct a secure face recognition system. The method will aid in the automatic recognition of faces while ensuring privacy. Citation: Hossain MM, Rahman MM (2026) Enhancing face recognition privacy through the integration of differential privacy and convolutional neural network. PLoS One 21(8): e0353565. https://doi.org/10.1371/journal.pone.0353565 Editor:
Aug 18, 2026 · via journals.plos.org
THE clock was ticking. I’d zipped round the supermarket, self-scanning as I went, and I now had about 15 minutes to exit, pack the car and get to work. But here was a security guard telling me everything would need to be scanned again. After unpacking the first bag, he lifted out an empty crisp packet. “Did you eat these while you were going round?” he asked. No, I replied, I was not eating crisps while shopping, and I had not shoplifted a packet of Walker’s Baked. He rummaged in the bag for a few other pieces of forgotten litter – old shopping lists and receipts – and offered them to me. “There’s a bin over there.” I stared at him. “You can put those right back where you found them,” I said. READ MORE: We work with refugees in Glasgow. The BBC film on our city had these glaring flaws Grumpily, he complied. It then transpired that in the process of re-scanning my entire weekly shop, he had somehow accidentally deleted the whole lot from the till, and from my app. The clock kept ticking. As I scanned it all again myself, I asked: “So what was it I missed?” He looked blank. “What was missing, when you checked the first 10 items?” “Oh, nothing. Sometimes we just have to do a full re-scan regardless.” Perhaps it was a coincidence that I had recently written in these pages about supermarkets displaying discounts the night before they come into effect – a clear breach of advertising rules – apparently to avoid having to pay workers for a night shift. Because of course I was neither a known thief nor an anonymous customer. Through my use of the self-scan app and loyalty scheme, the supermarket knows more about my tastes and
Aug 18, 2026 · via thenational.scot
Figures Abstract Large domain-specific foundation models have been widely adopted for retinal image analysis, yet systematic evidence for their advantage over compact general-purpose architectures remains scarce. We benchmarked nine model configurations spanning 22.8M to 303M parameters (vision transformers, hierarchical Swin Transformers, ConvNeXt, and the domain-specific RETFound models) across four tasks: OCT 8-class disease classification, and three fundus photography tasks (DME severity, glaucoma detection, and DR severity grading). All models were evaluated under identical training conditions, with both pretrained (on natural-domain image datasets) and from-scratch initializations compared using Mann-Whitney U tests. Pretraining improved accuracy by 5.18–18.41 percentage points across all tasks (p < 0.05 throughout), with larger benefits for CFP modalities and harder tasks. Compact hierarchical models (27–29M parameters) matched or exceeded larger architectures on three of four tasks. For instance, the SwinV2-tiny architecture ranked first on OCT, DME, and GL classification. The domain-specific RETFound model (303M) achieved the highest accuracy only on the most challenging task (DR severity grading, where the most severe class is underrepresented at 8% of images), where it outperformed the best compact model by 1.54 percentage points. These results indicate that compact general-purpose models may be sufficient for most retinal classification benchmarks, and that domain-specific foundation models may add higher value mainly for severity grading tasks with skewed class distributions. Citation: Isztl D, Spitznagel T, Somfai GM, Santos R (2026) Compact vision models match domain-specific foundation models for several retinal imaging classification tasks: A systematic benchmark. PLoS One 21(8): e0356202. https://doi.org/10.1371/journal.pone.0356202 Editor: Jia-Lang Xu, National Taichung University of Science and Technology, TAIWAN Received: April 14, 2026; Accepted: July 29, 2026; Published: August 18, 2026 Copyright: © 2026 Isztl et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium,
Aug 18, 2026 · via journals.plos.org
Nashville Metro Council considers new rules for police drones NASHVILLE, Tenn. (WZTV) — Metro Council is considering new rules that could change how Nashville police use drones as first responders, including when those drones can be deployed and what technology they can use. The debate centers on a proposal originally filed by Councilmembers Sandra Sepulveda and Ginny Welsch, a substitute version of that proposal, and a separate amendment that would add special events to the situations covered by the emergency-use rules. At the center of the discussion is MNPD's Drone as First Responder, or DFR, program. What is a Drone as First Responder? MNPD began a limited DFR trial in May 2026. The department says the program uses small drones that can respond to police and fire calls and, in some cases, arrive before officers. The goal is to give responders an aerial view of what is happening so they can make safer and more informed decisions. The initial trial uses three drones launched from the roof of the Madison Precinct. MNPD said the drones could respond within a two-mile radius to emergency calls, active criminal investigations, missing-person calls and significant traffic crashes. Four FAA-certified officers operate the drones from the Community Safety Center at police headquarters. MNPD says the program is not intended to be a general surveillance or routine patrol system. The department says drones are dispatched to specific calls for service. The department has also pointed to early examples of the technology being used during active calls. In May, MNPD said a DFR drone located a domestic-violence suspect who had fled on foot. In July, the department said drone footage helped officers arrest two suspects in an active robbery and recover a boxcutter used in the crime. PAST COVERAGE | Metro review board raises questions about MNPD
Aug 18, 2026 · via fox17.com