Yes, their neural-net based image tuning is the core IP/feature for the vision part of the product.
The comparison images/videos look like their ISP tuning (/ neural net) is cranking up the saturation too much and metering only in the center. Might just be the way it was set for the promotional video, so time will tell. All of these can be easily integrated into Macbook webcameras via software updates/tools.
"We gave the Opal C1 the fastest neural processing chip ever on a webcam. At 4 trillion operations per second, it rips through ML models at blazing speeds"
If they want/wanted optimal lens+sensor performance, the industrial designers would need to give a bit. The 0.9" depth is tricky for all-glass wide-angle lenses and 1/2.3" format type sensors. My guess is the device would need to be ~1.4" to incorporate an all-glass lens that's fully suitable for this sensor, particularly at F/1.8. Can talk more about this offline (commonlands.com).
It sounds like they are still figuring out their lens. The writing on the camera says 3.3mm~=100deg DFoV; 19mm equivalent on 35mm full frame. The webpage indicates 4.81mm~=78.6deg DFoV; 28mm equivalent on 35mm full frame.
Very clear the writer hasn't taken a course on optics:
"Rather than follow the curve of our eyes, the sensors in conventional digital cameras are flat. This results in an unnatural curvature in the image, and so the accompanying lenses have to be made in a way that corrects this distortion."
.... The aberration that curved image sensors helps mitigate is spherical aberration. Not distortion ....
Better, cheaper, and/or smaller lenses if the design is made specifically for the curved sensor.
TLDR; get used to using a phone case to hide the camera bump.
Low-light is the primary competitive motivation for mobile phone manufacturers. Most metasurfaces have >25% transmission loss per surface when manufactured.
This is a cool research concept. >20 years away and doubtful it'll go into the mobile phone rear-facing cameras. Maybe the structured light forward facing cameras will use it.
So, to remove the 'camera bump' they'd need to design an F/1.2 lens to be equivalent to current F/1.6 iPhone 1X lenses. That assumes they can achieve comparable MTF with only 2 metasurfaces. If they have to use 3 surfaces, it extends to F/1.0. You can do the math to see how quickly it falls apart using this calculator: https://commonlands.com/pages/dof-calculator .
The comparison images/videos look like their ISP tuning (/ neural net) is cranking up the saturation too much and metering only in the center. Might just be the way it was set for the promotional video, so time will tell. All of these can be easily integrated into Macbook webcameras via software updates/tools.
"We gave the Opal C1 the fastest neural processing chip ever on a webcam. At 4 trillion operations per second, it rips through ML models at blazing speeds"
If they want/wanted optimal lens+sensor performance, the industrial designers would need to give a bit. The 0.9" depth is tricky for all-glass wide-angle lenses and 1/2.3" format type sensors. My guess is the device would need to be ~1.4" to incorporate an all-glass lens that's fully suitable for this sensor, particularly at F/1.8. Can talk more about this offline (commonlands.com).
It sounds like they are still figuring out their lens. The writing on the camera says 3.3mm~=100deg DFoV; 19mm equivalent on 35mm full frame. The webpage indicates 4.81mm~=78.6deg DFoV; 28mm equivalent on 35mm full frame.