3 ms·

Eh. Modern phone cameras produce a kind of synthesized picture composed by machine learning, based on what you point at. That seems to be the direction of trave

Eh. Modern phone cameras produce a kind of synthesized picture composed by machine learning, based on what you point at. That seems to be the direction of travel. E.g. super-resolution based on multiple samples over time and integrated backwards to eliminate blur.

rabuse 6y ago
I honestly want the large camera manufacturers to start tapping into the computational photography realm. That tech, with their amazing sensors and glass, would produce some amazing results.
- jdhn 6y ago
  There was a post here on HN a few months ago where someone went into how the camera manufacturers are going to be challenged by phone manufacturers coming up with better and better algorithms to enhance the quality of pictures that are bumping up against physical limits such as lens size and pixel density. Edit: Here's [0] the link to the article. [0] https://photographylife.com/smartphone-vs-camera-industry https://photographylife.com/smartphone-vs-camera-industry
- nradov 6y ago
  It doesn't have to be in the camera. You can do that in post with the raw files.
  - crazygringo 6y ago
    Generally no, because a RAW file records only the total amount of exposure per "pixel". Full computational photography requires exposure per pixel over time (so something like a "stacked RAW" needs to be invented) combined with synced data from accelerometer/gyroscope/etc. for deconvolution.
- dingaling 6y ago
  I honestly don't. I wish they'd spend less time developing increasingly bloated and buggy camera software features and more time on simple improvements ( like exposure bracketing using ISO instead of just shutter speed )
Balgair 6y ago
It's still a '99% of people' solution. For the remaining people that really do want/need that last 1%, ML doesn't work. You can't infer data that does not exist no matter how many samples you take. The resolution of every object is set by the optical resolution: a function of the wavelength and the NA of the lens [0]. Aside: super-resolution refers to nanoscopes that break the Abbe' diffraction limit of light [1]. The various set-ups do this in a few very clever ways so the specifics change exactly how well they resolve objects. [0] https://en.wikipedia.org/wiki/Optical_resolution https://en.wikipedia.org/wiki/Optical_resolution [1] https://www.microscopyu.com/techniques/super-resolution/the-diffraction-barrier-in-optical-microscopy https://www.microscopyu.com/techniques/super-resolution/the-... (a good intro to these nanoscopes)

rabuse 6y ago

I honestly want the large camera manufacturers to start tapping into the computational photography realm. That tech, with their amazing sensors and glass, would produce some amazing results.

jdhn 6y ago
There was a post here on HN a few months ago where someone went into how the camera manufacturers are going to be challenged by phone manufacturers coming up with better and better algorithms to enhance the quality of pictures that are bumping up against physical limits such as lens size and pixel density. Edit: Here's [0] the link to the article. [0] https://photographylife.com/smartphone-vs-camera-industry https://photographylife.com/smartphone-vs-camera-industry
nradov 6y ago
It doesn't have to be in the camera. You can do that in post with the raw files.
- crazygringo 6y ago
  Generally no, because a RAW file records only the total amount of exposure per "pixel". Full computational photography requires exposure per pixel over time (so something like a "stacked RAW" needs to be invented) combined with synced data from accelerometer/gyroscope/etc. for deconvolution.
dingaling 6y ago
I honestly don't. I wish they'd spend less time developing increasingly bloated and buggy camera software features and more time on simple improvements ( like exposure bracketing using ISO instead of just shutter speed )

Balgair 6y ago

It's still a '99% of people' solution. For the remaining people that really do want/need that last 1%, ML doesn't work. You can't infer data that does not exist no matter how many samples you take. The resolution of every object is set by the optical resolution: a function of the wavelength and the NA of the lens [0]. Aside: super-resolution refers to nanoscopes that break the Abbe' diffraction limit of light [1]. The various set-ups do this in a few very clever ways so the specifics change exactly how well they resolve objects. [0] https://en.wikipedia.org/wiki/Optical_resolution https://en.wikipedia.org/wiki/Optical_resolution [1] https://www.microscopyu.com/techniques/super-resolution/the-diffraction-barrier-in-optical-microscopy https://www.microscopyu.com/techniques/super-resolution/the-... (a good intro to these nanoscopes)