4 ms·

Why isn't depth a way bigger deal? Cameras could capture depth and give you whatever virtual lens you want, in camera like a phone filter. Lenses are fun and c

Why isn't depth a way bigger deal? Cameras could capture depth and give you whatever virtual lens you want, in camera like a phone filter. Lenses are fun and collectable and all, but I think my idea of a near perfect camera would be a one piece waterproof superzoom with depth capture and AI processing. Less to fuss with, and you can carry all your virtual lenses everywhere.

radlad 4y ago
> AI processing This one is super interesting to me - clearly Snapchat, Google, Apple, and I assume Samsung have proved that this is viable. While some features (e.g. portrait mode) exist to work around limitations that many lenses don't have (ability to create a shallow DoF) other ones are more interesting, i.e. Snapchat filters. Still, I worry that these are too difficult to reason about as compared to something like the exposure triangle, and the imperfections of things like portrait mode can be very obvious. Perhaps these are better left to post-processing tools?
- mgdlbp 4y ago
  In-camera-phone processing has the advantage of access to depth and accelerometer data recorded during potentially multiple sequential exposures of multiple cameras for stacking and motion blur deconvolution. There's a parallel in the advent of stacking being a massive boon for amateur astrophotography. There, the natural format for unprocessed data is video, hours of which are recorded in-camera and processed afterwards. Perhaps someday a new RAW-over-time format will be developed to hold the raw data of multiple sensors, optical and otherwise, to enable smartphone processing after the fact.
  - retcon 4y ago
    Spot on about the new picture being a stream. "Super resolution" may be the marketing term to search for to investigate your interest. However sequential image composite decononvolution hasn't solved for usually desired out of focus transition effects commonly hand waved as " bokeh" because the important contrast gradient limit of the Airy Disc is convolution. Accelerometers are appearing even in the lowest budget cinema cameras whilst ARRI, the de facto market leader who used to be the left field upstart (finishing The Bridge Over The River Kwai, critically being inexpensive light and ergonomic of which only ergonomic is arguably retained by contemporary boxes) has just gotten around to putting one in a smaller sensor camera that costs six times the bigger sensor competition. (I'll ignore the 3:2 vs 4:3 ratio of the newest ARRI is wonderful for anamorphic squeeze and utterly wasted for spherical shooting.. and ignore RED offering very usable ratio comprises. Edit: disc not discussion
solardev 4y ago
Lightfield cameras were briefly a thing, but then I guess the engineers at Google and Apple figured they could fake it well enough with algorithms + depth sensors. https://en.wikipedia.org/wiki/Light_field_camera https://en.wikipedia.org/wiki/Light_field_camera It would be cool to see a traditional SLR/mirrorless with a depth sensor, indeed!
- mgdlbp 4y ago
  The pinnacle of the multiple camera trend: spanning the entire back surface of a phone, a single light field sensor!
- orbital-decay 4y ago
  Smartphones had nothing to do with Lytro's death, though; DSLR users ultimately decided that they liked super sharp pictures more than the ability to focus in post.
  - aspyct 4y ago
    Not sure how Lytro worked. Was it a one-shot, fast enough to freeze action? Otherwise there's what we call focus bracketing/stacking which can also allow us to focus in prod. However, there's usually only one good focus point, so we can do that while taking the picture...
isatty 4y ago
Depth processing and the portrait modes on both the iPhone and pixel is pretty bad though, if you know what real glass can do. For a simple example, they still mess up the whiskers on a kitty and even for portraits of people it’s super obvious that it’s an artificial image.

radlad 4y ago

> AI processing This one is super interesting to me - clearly Snapchat, Google, Apple, and I assume Samsung have proved that this is viable. While some features (e.g. portrait mode) exist to work around limitations that many lenses don't have (ability to create a shallow DoF) other ones are more interesting, i.e. Snapchat filters. Still, I worry that these are too difficult to reason about as compared to something like the exposure triangle, and the imperfections of things like portrait mode can be very obvious. Perhaps these are better left to post-processing tools?

mgdlbp 4y ago
In-camera-phone processing has the advantage of access to depth and accelerometer data recorded during potentially multiple sequential exposures of multiple cameras for stacking and motion blur deconvolution. There's a parallel in the advent of stacking being a massive boon for amateur astrophotography. There, the natural format for unprocessed data is video, hours of which are recorded in-camera and processed afterwards. Perhaps someday a new RAW-over-time format will be developed to hold the raw data of multiple sensors, optical and otherwise, to enable smartphone processing after the fact.
- retcon 4y ago
  Spot on about the new picture being a stream. "Super resolution" may be the marketing term to search for to investigate your interest. However sequential image composite decononvolution hasn't solved for usually desired out of focus transition effects commonly hand waved as " bokeh" because the important contrast gradient limit of the Airy Disc is convolution. Accelerometers are appearing even in the lowest budget cinema cameras whilst ARRI, the de facto market leader who used to be the left field upstart (finishing The Bridge Over The River Kwai, critically being inexpensive light and ergonomic of which only ergonomic is arguably retained by contemporary boxes) has just gotten around to putting one in a smaller sensor camera that costs six times the bigger sensor competition. (I'll ignore the 3:2 vs 4:3 ratio of the newest ARRI is wonderful for anamorphic squeeze and utterly wasted for spherical shooting.. and ignore RED offering very usable ratio comprises. Edit: disc not discussion

solardev 4y ago

Lightfield cameras were briefly a thing, but then I guess the engineers at Google and Apple figured they could fake it well enough with algorithms + depth sensors. https://en.wikipedia.org/wiki/Light_field_camera https://en.wikipedia.org/wiki/Light_field_camera It would be cool to see a traditional SLR/mirrorless with a depth sensor, indeed!

mgdlbp 4y ago
The pinnacle of the multiple camera trend: spanning the entire back surface of a phone, a single light field sensor!
orbital-decay 4y ago
Smartphones had nothing to do with Lytro's death, though; DSLR users ultimately decided that they liked super sharp pictures more than the ability to focus in post.
- aspyct 4y ago
  Not sure how Lytro worked. Was it a one-shot, fast enough to freeze action? Otherwise there's what we call focus bracketing/stacking which can also allow us to focus in prod. However, there's usually only one good focus point, so we can do that while taking the picture...

isatty 4y ago

Depth processing and the portrait modes on both the iPhone and pixel is pretty bad though, if you know what real glass can do. For a simple example, they still mess up the whiskers on a kitty and even for portraits of people it’s super obvious that it’s an artificial image.