Topics
Latest
AI
Amazon
Image Credits:Glass
Apps
Biotech & Health
Climate
Image Credits:Glass
Cloud Computing
commercialism
Crypto
Concept animation showing process of going from RAW to Glass-processed image.Image Credits:Glass
Enterprise
EVs
Fintech
Photo provided by Ziv Attar showing an iPhone 15 Pro Max zoomed to 5x, and the Glass-processed version of the phone’s RAW images.Image Credits:Ziv Attar
fund raise
gadget
Gaming
Image Credits:Glass / DXO
Government & Policy
Hardware
Left: RAW, right: Glass-processed.Image Credits:Glass
Layoffs
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
surety
Social
infinite
inauguration
TikTok
fare
speculation
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
television
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
Your phone ’s photographic camera is as much software as it is computer hardware , and Glass is hoping to better both . But while itswild anamorphic lenscreeps to market , the company ( hunt on $ 9.3 million in new money ) has released an AI - power tv camera upgrade that it says vastly improves image quality — without any weird AI upscaling artifacts .
GlassAI is a purely software system plan of attack to improving epitome , what they call a neural image signal processor ( ISP ) . ISPs are basically what take the raw sensing element output — often unconditional , noisy and distorted — and turn that into the sharp , colorful image we see .
The ISP is also increasingly complex , as phone manufacturing business like Apple and Google care to show , synthesize multiple exposures , quickly find and sharpening faces , adjusting for tiny movements , and so on . And while many include some shape of machine learnedness or AI , they have to be thrifty : Using AI to generate detail can create hallucinations or artifacts as the system tries to make visual information where none exists . Such “ super - firmness ” model are useful in their space , but they have to be carefully monitored .
Glass make both a full camera system base on an unusual pill - shaped front element , and an ISP to back it up . And while the former is working toward grocery presence with some coming equipment , the latter is , it turns out , a product deserving sell in its own right .
Glass rethinks the smartphone camera through an onetime - school movie theatre lens
“ Our restoration networks correct optic aberration and sensing element number while efficiently take away dissonance , and surpass traditional Image Signal Processing pipelines at all right texture recuperation , ” explained CTO and carbon monoxide gas - laminitis Tom Bishop in their news firing .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
The parole “ recuperation ” is key , because contingent are not only created butextractedfrom raw mental imagery . bet on how your camera stack already put to work , you may bed that certain artefact or angles or noise form can be reliably resolve or even take advantage of . instruct how to reverse these entail details into actual ones — or conflate item from multiple exposures — is a giving part of any computational photography batch . Co - founder and CEO Ziv Attar says their nervous ISP is better than any in the industry .
Even Apple , he pointed out , does n’t have a full neuronal image stack , only using it in specific circumstances where it ’s needed , and their results ( in his notion ) are n’t neat . He provided an example of Apple ’s neural ISP bomb to interpret schoolbook correctly , with Glass faring much good :
“ I think it ’s middling to adopt that if Apple has n’t manage to get comely results , it is a concentrated job to solve , ” he said . “ It ’s less about the actual plenty but more about how you train . We have a very unique path of doing it , which was developed for the anamorphic lense systems and is efficient at any television camera . fundamentally , we have breeding labs that necessitate robotics systems and optical standardisation system that manage to prepare a meshwork to characterize the aberration of genus Lens in a very comprehensive way , and fundamentally reverse any optical distortion . ”
As an example , he provided a case cogitation where they had DXO judge the camera on a Moto Edge 40 , then do so again with GlassAI put in . The Glass - processed images are all clear improved , sometimes dramatically so .
At humble light storey the built - in ISP struggles to differentiate fine lines , textures and facial details in its dark style . Using GlassAI , it ’s as sharp as a tack even with half the vulnerability time .
you could go peep the pixelson a few test photos Glass has availableby switching between the birthday suit and the final .
Companies putting together phones and cameras have to drop a band of time tuning the ISP so that the sensing element , lens and other scrap and pieces all play together properly to make the good image potential . It seems , however , that Glass ’s one - size - fit - all operation might do a better Book of Job in a fraction of the clip .
“ The prison term it take us to train shippable software from the time we put our manus on a young character of gadget … it depart between few 60 minutes to few days . For denotation , phone makers spend calendar month tuning for image quality , with immense teams . Our process is amply automated so we can support multiple devices in a few day , ” said Attar .
The neural ISP is also end - to - end , meaning in this context that it goes direct from sensor RAW to final image with no extra processes like denoising , sharpening and so on needed .
When I asked , Attar was careful to differentiate their oeuvre from A-one - resolution AI services , which take a finished image and upscale it . These often are n’t “ reclaim ” details so much as inventing them where it seems appropriate , a process that can sometimes get unsuitable results . Though Glass apply AI , it is n’t reproductive the way many image - colligate AIs are .
Today nock the product ’s accessibility at prominent , presumably after a lengthy examination period with partners . If you make an Android telephone , it might be sound to at least give it a shot .
On the computer hardware side , the telephone with the unearthly lozenge - shape anamorphic camera will have to await until that manufacturer is ready to go public , though .
While Glass develops its tech and is trying out customers , it ’s also been busy scaring up funding . The company just closed a $ 9.3 million “ extended seed , ” which I put in quotes because the seed stave was in 2021 . The unexampled funding was led by GV , with Future Ventures , Abstract Ventures and LDV Capital participating .
