Posts

Showing posts with the label machine vision

Spatiotemporal Photonic Interleaving Network enables high-density optical convolutional processing for light-speed computing

https://www.eurekalert.org/news-releases/1138638 "chip-area optical processor bottleneck overcome routing light across recursive tree of cascaded shared wavelength domain delay segments... scalable, reconfigurable, reduces waveguide-length complexity... validated using 220-nm silicon-on-insulator chip operating at 49 GBd, processed handwritten digits with >0.98 optical-to-digital output agreement... enables multiple image-processing tasks/ configurable kernel geometries to share same core... signal processing projected to reach 29.7 trillion operations/ second"
Post a Comment
Read more

Cluster-aware upcycling expands pretrained foundation models into specialized mixture-of-experts without retraining from scratch

https://www.eurekalert.org/news-releases/1139465 "upcycling machine vision learning duplicates identical weights or assigns routing randomly, overcome extracting underlying semantic structure from token activations in pretrained layers grouping similar features into distinct clusters from which expert weight matrices/ routing centroids initialized... outperforms standard sparse upcycling across image-text retrieval/ multiple visual classifications... scalable, preserves existing AI assets, reduces computational overhead... large-scale multimodal AI"
Post a Comment
Read more

Self-powered photodiodes integrating sub-20 nm ferroelectric layer into WSe2 channel with sub-fJ programming energy for edge computing

https://www.eurekalert.org/news-releases/1138477 "vision processing has latency/ energy bottlenecks from image sensing/ computing separation... overcome with ferroelectric homojunction using reconfigurable photodiode integrating: ambipolar tungsten diselenide/ hafnium zirconium oxide... split-gate enables polarity-switchable photocurrent under 0 external bias, allowing optical signal processing at pixel level... 50 Î¼s switching speed, >100 seconds non-volatile state retention... CMOS-compatible energy-efficient unified sensing/ memory/ computing vision systems"
Post a Comment
Read more

Upsample Anything optimizes computer vision models, delivering high-resolution outputs while slashing GPU memory allocation 16X

https://techxplore.com/news/2026-06-upsampling-method-sharpens-ai-vision.html "vision models compress high-resolution images to conserve memory, but discards small objects/ fine structures... overcome utilizing test-time optimization/ Joint Bilateral Upsampling dynamically restoring compressed features back to original resolution during inference... deploys high-fidelity computer vision without heavy hardware overhead... training-free algorithm: efficient, cost-saving, scalable: resource-constrained AI edge devices, smartphones, autonomous robots, enterprise data centers"
Post a Comment
Read more

Biomimetic titanium dioxide photomemristor vision system adapts to extreme, fluctuating, and mixed lighting conditions in seconds

https://www.eurekalert.org/news-releases/1130994 "computer vision in high-contrast/ rapidly shifting lighting, such as dark nighttime street illuminated by oncoming high-beam headlights, unsafe for autonomous transport, overcome using titanium dioxide/ flexible conductive polymer blend photomemristor... absorbs/ releases surrounding moisture based on environmental light exposure, regulating optical sensitivity within seconds... 4X4 sensor grid/ machine learning identified complex letter patterns in high-contrast environments with over 95% accuracy"
Post a Comment
Read more

Matte/ shiny differentiation by machine vision enhances robotic surgery, industrial inspection, autonomous vehicle glare navigation

https://www.eurekalert.org/news-releases/1128523 "3D sensors can't accurately map mix of matte surfaces (like fabrics, walls)/ highly reflective, specular surfaces (like glass, mirrors, wet surgical tissue)... overcome using deflectometry: algorithm/ laser scanner instantly maps dull, matte object 3D shapes/ repurposes as collective virtual screen, then tracks how light reflects off shiny objects onto this surrounding virtual screen, computing precise shapes of reflective surfaces... moving, mixed-reflectance scene, real time high-resolution 3D video... highly scalable"
Post a Comment
Read more

Artificial optoelectronic synapse with >25 minutes of continuous memory retention vastly simplifies circuitry/ cuts energy consumption

https://phys.org/news/2026-05-optoelectronic-synapse-exceptional-photoresponse-neuromorphic.html "utilizes solution-processed mesoscopic α-phase vanadium pentoxide thin films... manipulates oxygen vacancies within crystalline lattice enabling persistent photoconductivity functioning intrinsic, gate-free memory... broadband light signal captures/ accurately mimics biological eye-brain pathways, establishing short/ long-term synaptic plasticity... exceptional stability/ retention over 150 bending cycles... optical sensing/ nonvolatile memory integration... autonomous: edge-AI vision, machine vision" Related: Photon-driven synapse advances low-power neuromorphic systems https://techxplore.com/news/2026-05-photon-driven-synapse-advances-power.html Stretchable brain-inspired electronics erase the physical boundary between human and machine https://www.eurekalert.org/news-releases/1130860 Brain-inspired nanopore device uses current-induced heating for memory operations https://phys...
Post a Comment
Read more

Tackling optical neural networks' interference enables efficient, low power, speed of light diffractive edge-AI multi-object recognition

https://www.eurekalert.org/news-releases/1120604 "computers using structured light diffraction through layered materials have interference/ cross-talk among multiple objects, better design: reduces cross-talk between different objects, improves simultaneously present multiple target robustness, optimizes how light propagates/ processed across diffractive layers engineered to encode features of different objects... processing happens via light propagation (no electronics required during inference)... autonomous systems, surveillance, biomedical imaging" Related: The quantum key to seeing through chaos https://phys.org/news/2026-05-quantum-key-chaos.html
Post a Comment
Read more

Summit, the world's first long-range, high-resolution THz vision sensor, combines lidar's high resolution with radar's weather-resistance

https://interestingengineering.com/ces-2026/worlds-first-terahertz-vision-sensor "984 feet range, 0.13° native resolution (up to 20X better than current automotive radar), provides detailed 3D point clouds with Doppler data (range, direction, elevation, speed), solid-state digital phased array using proprietary modular THz engine... unlike cameras or lidar, THz waves can see through fog, rain, snow, sleet, sun glare without losing accuracy... advanced driver-assistance system/  L2 to L5 autonomous driving levels... high-volume production 2028... motor vehicles, defense, healthcare, manufacturing" Related: Extra set of eyes for self-driving cars: Roadside radar sensors could reduce blind spots https://www.eurekalert.org/news-releases/1118398 A smarter way to debias AI vision models https://techxplore.com/news/2026-04-whac-mole-dilemma-smarter-debias.html Light-tunable polarization sensor could sharpen self-driving cars and medical scans https://techxplore.com/news/2026-05-tuna...
Post a Comment
Read more

Cross-View Cross-Modality Correspondence, an AI model, allows computers to bridge the gap between real-world photographs and 2D floor plans

https://techxplore.com/news/2025-12-vision-method-links-photos-floor.html "maps/ photos look nothing alike, making correlating by AI difficult, overcome with dataset of 90K pairs floor plans/ photos (manually aligned) across nearly 600 internet photos turned into 3D models... includes 153 million pixel-level matches, allowing AI to see spot in photo corresponding to map coordinate... 34% reduced errors... step toward multi-modal 3D computer vision... robots navigating indoor spaces using only blueprint, AR help with way through buildings, digital modeling: 3D architecture reconstructions" Related: A human-inspired pipeline could enhance the training of computer vision models https://techxplore.com/news/2026-05-human-pipeline-vision.html
Post a Comment
Read more

Microelectronics fabrication that significantly improves energy efficiency by stacking active components on the back end of a computer chip

https://www.eurekalert.org/news-releases/1109566 "logic (transistors)/ memory fabricated in 1 compact stack on 1 chip... 2 nm thick amorphous indium oxide active channel layer grown at relatively cool 150°C preventing damage to existing front-end circuitry, minimized defects (oxygen vacancies)... added ferroelectric hafnium-zirconium-oxide layer, ending up with 20 nm back-end transistors with built-in memory, 10 ns switching speed, lower voltage... generative AI, deep learning, computer vision... future: integrate these back-end memory transistors onto single circuit"
Post a Comment
Read more

Spectral Kernel Machines' single semiconductor integrates: machine-learning inference during photodetection, spectral imaging, machine vision, AI

https://techxplore.com/news/2025-11-intelligent-photodetectors-dogs-materials-spectra.html "directly encodes visible to mid-IR spectral/ spatial information into output photocurrent, so sensor identifies materials, chemicals,.. optoelectronic sensors learn/ compute without need for digital post-processing... electrically tunable enhancing/ suppressing specific spectral signatures... trains sampling pixels/ receiving simple commands, learns optimal electrical control sequence targeting object's pixels/ suppress background... for new/ unseen images, produces only for pixels belonging to target object" Related: Bioinspired robot eye adjusts its pupil to handle harsh lighting https://techxplore.com/news/2026-03-bioinspired-robot-eye-adjusts-pupil.html New chip lets robots see in 4d by tracking distance and speed simultaneously https://www.nature.com/articles/d41586-026-00727-1 Meta-optics enabling highly scalable machine vision https://www.eurekalert.org/news-releases/1120546...
Post a Comment
Read more

Low energy consumption photoelectric memristors with multi‑level linear conductance modulation for slimmer, smarter artificial visual systems

https://www.eurekalert.org/news-releases/1097964 "AI chip on-sensor vision acts like optical synapse... sense, store, process information retina-like... single 0.5 V, 5 ms optical pulse consumes only 187 pJ, conductance updates show 99.6 % linearity eliminating redundant weight tuning/ cutting system power... interface traps photo-carriers, dynamically lowering barrier/ enabling multi-level linear conductance modulation, 2-terminal cross-point structure, wide-band response allows outdoor/ night-time operation without filters... autonomous night driving" Related: From Biology to Electronics: BIOPIX, the Artificial Retina Born at Tor Vergata https://web.uniroma2.it/en/contenuto/BIOPIX-artificial-retina-born-at-tor-vergata
Post a Comment
Read more

Low energy consumption photoelectric memristors with multi‑level linear conductance modulation in slimmer/ smarter AI on-sensor vision

https://www.eurekalert.org/news-releases/1097964 "single chip senses, stores, processes... single 0.5 V, 5 ms optical pulse consumes 187 pJ... 99.6 % linearity in learning eliminates redundant weight tuning/ reduces system power... optical retina/ synapse -like functions: paired-pulse facilitation, long-term plasticity/ learning reproduced under visible light... facial recognition, autonomous night driving... integrates sense-memory-compute pixels into bionic eyes/ smart cameras/ neuromorphic edge devices... future: new oxide stacks for even lower energy/ higher speed"
Post a Comment
Read more

Artificial neuron merges DRAM/ MoS₂ circuits emulating synaptic plasticity for energy-efficient machine learning, computer vision/ image recognition

https://techxplore.com/news/2025-08-artificial-neuron-merges-dram-mos.html "combines dynamic random-access memory with ultrathin semiconductor monolayer molybdenum disulfide-based circuits creating integrate-and-fire neuron... DRAM stores electrical charges in capacitors, electrical charge amounts can be modulated to mimic variations in electrical charge across biological neurons' membrane determining whether inverter triggers electrical bursts... resembling those observed in biological neurons when they fire... also emulates photopic/ scotopic adaptation of human vision... edge computing" Related: A metal-organic framework neuron with dopamine perception https://www.eurekalert.org/news-releases/1097074
Post a Comment
Read more

Instead of producing images/ videos from scene descriptions, extract scene descriptions from simulated images using inverse generative AI

https://techxplore.com/news/2025-08-ai-method-reconstructs-3d-scene.html#google_vignette "generalizes well across new images or different scenarios/ extracted patterns traditionally difficult to interpret... gradually adjusts model's internal parameters... solve vision tasks, such as tracking, as test-time optimization problems... new datasets not needed for training on... compares rendered image with real observed image, backpropagating difference through differentiable rendering function/ 3D generation model to update its inputs... provides 3D explanations of its perceived world... robotics, autonomous driving"
Post a Comment
Read more

Ionic-electronic photodetector performs in-sensor image processing, offering the potential to surpass color vision deficiencies of human vision

https://phys.org/news/2025-08-ionic-electronic-photodetector-sensor-vision.html "van der Waals ferroelectric layered copper indium phosphorus sulfide photodetector supports ionic/ electronic conduction... mobile Cu+ ion motion elicits nonlinear/ history-dependent photoresponses, light sensitivity dynamic tuning... reconfigurability allows selectively enhancing weak signals/ suppressing overexposed regions... reweight spectral contrast/ modulate responsivity enhancing color contrast between red/ green...noise removal, contrast enhancement, image inversion filtering, low-power edge AI"
Post a Comment
Read more

Wearable bionic device mimics birds' extraordinary sight and uses almost no power, enabling more efficient and advanced machine vision

https://techxplore.com/news/2025-02-wearable-bionic-device-mimics-birds.html "quickly identifying/ classifying objects... combines advanced materials with neural networks... low-light condition capable, broadband non-volatile storage capabilities... seamlessly combines AI hardware with algorithms... specially oriented GaAs nanowire arrays combined with liquid-surface-assembled P3HT organic films to form van der Waals heterojunctions.... using reservoir computing to recognize moving object's shape, motion, color, UV grayscale... autonomous vehicles, robotics" Related: Neuromorphic system enhances machine vision in extreme lighting environments https://techxplore.com/news/2025-03-neuromorphic-machine-vision-extreme-environments.html Robotic eyes mimic human vision for superfast response to extreme lighting https://www.eurekalert.org/news-releases/1088968 New retina-inspired photodiodes could advance machine vision https://phys.org/news/2025-08-retina-photodiodes-advance-mac...
Post a Comment
Read more