Articles tagged "research"

NVIDIA Research creates AI tool to convert series of 2D images into impressive 3D models

Researchers at NVIDIA have created a new inverse rendering pipeline, 3D MoMa. It turns a series of images of a 2D object into a 3D object built upon a triangular mesh, allowing it to be used with a wide range of modeling tools and engines.

Google's Imagen AI produces photorealistic images from natural text with frightening fidelity

Artificial intelligence is improving fast. Less than a month after OpenAI released its impressive DALL-E 2 text-to-image generator, Google has bested it with Imagen.

Newly-developed lensless camera uses neural network and transformer to produce sharper images faster

Lensless cameras have many potential use-cases but have generally been held back by lengthy processing requirements and low-resolution images. A research from a team at the Tokyo Institute of Technology is looking to change that.

This light field camera inspired by an extinct trilobite promises high-resolution image with extreme depth of field

Focusing on an object a few centimeters in front of your camera is easy with a macro lens. What if you want to focus on something over a kilometer away simultaneously? New research shows that it's possible thanks to a light field camera with an incredible lens design inspired by an extinct arthropod's unique compound eyes.

Research shows how AI may help create full-color night vision cameras

Scientists at the University of California Irvine have shown how AI can be used to predict full-color images from infrared cameras.

Researchers develop an impressive style-based 3D-aware generator for high-res image synthesis

This novel approach integrates its neural radiance field (NeRF) into a style-based generator to improve rendering efficiency, resolution and 3D consistency compared to previous models.

Recapping the latest NVIDIA Research: turning 2D images into 3D models and a new NVIDIA Canvas update

NVIDIA has been busy. Last year, the company unveiled some fascinating AI-powered research projects, including technology for converting text into images and 2D images into 3D models. At CES last week, NVIDIA updated one of its most interesting AI tools and showed off a lot of new, powerful hardware.

Enhance! Google researchers detail new method for upscaling low-resolution images with impressive results

Google might just be turning CSI's laughable 'Enhance!' scenes into a reality.

Researchers propose ‘spaceplates’ to miniaturize lenses by reducing air gaps

A team from the University of Ottawa has shown that lenses can be made shorter when spaceplates made of metamaterials are introduced to the optical path. The findings promise miniaturized lenses with no loss of quality

Cornell University researchers break their record with stunning high-res photo of atoms

In 2018, using ptychography and sophisticated algorithms, a Cornell University research team set a record with an incredible photo of atoms. With improved detection and algorithms, the team is back with another record-setting image, and it's the highest-resolution photo of atoms ever captured.

MIT and UMass researchers develop world's first flat ultra-wide-angle fisheye lens

The lens is made from a single piece of thin glass, opening the door for a variety of potential new imaging applications from consumer tech to medical imaging.

Disney Research Studios demonstrates automatic face swapping with faster, cheaper AI

The study details a new algorithm that can produce high-resolution, photo-realistic face swaps, but head shape remains an issue.

Researchers develop lithium-sulphur battery that can power a phone for five days

Australian researchers have pioneered the development of a lithium-sulfur battery technology that may revolutionize consumer gadgets, cameras, electric cars and more.

Researchers have developed reset-counting pixel that promises near-limitless highlight capture

Researchers from Institut für Mikroelektronik Stuttgart have developed a pixel design with the potential for massively increased dynamic range thanks to the ability to 'count' the number of times an individual pixel resets when it becomes saturated with light.

Researchers use ‘bio-inspired’ event cameras and AI to generate ultra-high-speed videos

Researchers with the University of Zurich and ETH Zurich have shared a paper showing the development of a novel recurrent neural network that can reconstruct ultra-high-speed videos from data captured by event cameras.

NVIDIA Research project uses AI to instantly turn drawings into photorealistic images

GauGAN as it's known, can create photorealistic images from basic drawings using the power of artificial intelligence.

NVIDIA researchers create AI that generates photo-realistic portraits

NVIDIA researchers have created an AI that can create realistic portraits on demand using various elements from source images.

NVIDIA researchers can now turn 30fps video into 240fps slo-mo footage using AI

Researchers at NVIDIA have used artificial intelligence to train a system that's capable of turning standard 30fps video into 240fps slow-mo video with minimal loss in quality.

Study finds shooting and sharing one photo per day improves wellbeing

A study co-authored by researchers from the University of Sheffield and Lancaster University has found "complex benefits" associated with taking one image per day and sharing it with an online community.

Google just made the tech behind its 'portrait mode' open source

The 'semantic image segmentation model' categorizes every pixel in an image and assigns it a label, such as “road”, “sky”, “person” or “dog.” And now, Google has released its latest version as open source, making it available to any developers whose apps could benefit from the tech.

This 'metalens' breakthrough may revolutionize lenses as we know them

Researchers at Harvard have made a huge breakthrough in metalenses—flat 'lenses' that use nano structures to focus light—making it possible these flat lenses will some day replace the bulky, multiple-element lenses we use today.

Google's new AI ranks photos on their technical and aesthetic quality

Google's NIMA (Neural Image Assessment) model ranks photos on a scale of 1 to 10 in terms of technical and aesthetic quality. In testing, its rankings closely match scores assigned by human judges.

NVIDIA Computational Zoom lets you change perspective and focal length in post

Researchers with the University of California, Santa Barbara and NVIDIA have developed a new type of 'computational zoom' tech that will let you change a photo's perspective and focal length in post.

Stanford develops extra-wide 4D light field camera... for robots

Researchers at Stanford have revealed a new '4D camera system' built for robots. The system is based on the same light field tech that allowed Lytro cameras to refocus images after they were taken.

Study finds most people can't spot manipulated photos, can you?

How confident are you that you can spot a manipulated photo? A recent study at the University of Warwick shows that many people are pretty bad at it.

New research on bee vision might help improve digital cameras

New research on how bees perceive color might some day revolutionize how digital cameras judge ambient light conditions, producing more realistic and consistent colors.

Google software engineer shows what's possible with smartphone cameras in low light

Google software engineer Florian Kainz wrote a custom Android app and used manual post-processing in Photoshop to show what's possible with current smartphone cameras in low light. Read more

Researchers create method for photorealistic Prisma-style effects

Prisma, but photorealistic rather than painterly: researchers at Cornell and Adobe have presented a new method that uses deep learning to transfer the style of one photo to another. Read more

This CMOS sensor with 3D-printed microlenses is designed to mimic predator vision

Using 3D printing to create ultra-precise micro-lenses directly on a CMOS sensor might one day help drones see the way eagles and other predators do. Read more

Google AI adds detail to low-resolution images

A new system by Google based on neural networking is capable of adding detail to very low-resolution images. Read more

Google uses neural networks to improve image compression

A research team at Google has developed a way to use neural networks to compress image files in a more efficient way than current methods by building an artificial intelligence system and training it with random images from the internet. Read more

Microsoft's MobileFusion turns smartphones into 3D scanners

A Microsoft prototype app allows for the generation of 3D models using your smartphone.

Researchers in Tokyo develop high-speed subject tracking system

Engineers at Tokyo University's Ishikawa Oku Laboratory have come up with new technology to track extremely fast motion. Their new system - which uses 'Saccade Mirrors' for pitch and tilt, a 'pupil shift system', and very fast image processing - is able to keep even the quickest subject in the center of the frame at all times. According to engineers, the initial application for this system could be to capture video at sporting events. They expect it to be market-ready in about two years. Follow the link for a video demonstration of this intriguing new technology.

New study investigates online reviews - makes surprising discoveries

Online product reviews are a huge part of our lives these days, and often, they're the nearest thing we get to a 'hands on experience' before we buy something. But how do we know they're honest? Dishonest reviews are a serious problem for retailers (and indeed for content websites like and a team at the Massachusetts Institute of Technology has been investigating why people are driven to leave reviews of products that they haven't actually used. Their findings might just surprise you. Click through to read more. 

Engineers eye-up insect biology as inspiration for curved camera

Engineers have developed a curved camera designed to mimic insects' compound eyes. The hemispherical design gives a wide field-of-view with no aberrations and effectively infinite depth-of-field, with the hope it could be used in applications such as endoscopy or as visual sensors on unmanned aircraft. The current design uses 180 light-sensitive elements, each behind its own lens, but researchers hope to build one with 20,000 elements, giving a similar resolution to that seen by dragonflies.

Scientists demonstrate 'paint-on' batteries

Scientists in Texas have demonstrated a way of 'painting' rechargeable lithium-ion batteries onto surfaces, greatly expanding the potential for future development of portable electronics. The team, from Rice University, has succeeded in painting batteries onto a range of different surfaces, including common household objects, with 'no surface conditioning'. The batteries are made up of five layers measuring just 0.5mm thick in total and, according to the scientists that developed the technology, can be fabricated using conventional spray-painting equipment and techniques.

One-shot gigapixel camera offers a future beyond flat sensors

A gigapixel camera developed for the US Department of Defense's research agency provides an insight into the challenges that will need to be overcome to offer super-high-resultion cameras. A team from Duke University, Durham, North Carolina, has described its 960 megapixel (0.96 gigapixel) 'AWARE-2' camera in a letter to scientific journal Nature. The team says small, efficient electronics are the key to being able to miniaturize the camera, which currently sits in a 0.75 x 0.75 x 0.5m frame.

Scientists develop flexible sensor to allow simple zoom

Scientists have successfully constructed a digital camera that can be flexed to focus an image, allowing its use with simple single-element lenses. Researchers from Northwestern University and the University of Illinois at Urbana–Champaign created a 16 x 16 pixel array on an elastomeric backing that can be distorted to correctly focus the image from a simple lens. In a paper to be published in the Proceedings of the National Academy of Sciences (PNAS), they combine this with a single-element, tunable lens to provide a camera with very simple optics capable of zooming. The technology could eventually provide 'studio quality' images from cellphone cameras, one of the lead researchers says.