It is a sort of geometric optics approximation as I understand, otherwise it was an electromagnetic or quantum field. It looks more for entertainment than for research purposes.
It's a better approximation than you can make with a normal camera, so its applications are even broader. Instead of a normal camera image which captures the position of focused light rays on a plane, this captures the position and direction of light rays as they move through a volume. Personally I would not call such a versatile instrument useless.
One of the best and earliest uses of VR, long before Vive or Oculus or the rest came around, was for treating vets with PTSD. The technology was more or less indistinguishable from entertainment, apart from its application.
Point being, don’t judge this without knowing more.
I just played their demo. Even on a VR rig, this is one of the more resource intensive experiences.
Definitely very cool; cooler than it sounds. You get to experience a summation of all the subtle touches missing from VR today - like how a reflection on something changes as you rotate your head.
But like with all stuff VR, to "get it" you really have to experience it in VR. Especially how different this feels from 360 pics (stereoscopic or not) or photogrammetrie isn't obvious I think if you haven't seen it yourself.
My understanding is that there is an insane amount of data stored in the light field captures that must be streamed seamlessly to the headset. Can someone shed some light (heh) on the technical issues surrounding light fields?
Light fields are large, but not insane, because they are highly compressible. Think of them as a matrix of images, each representing a micro lens at a point in the field. Most patches of the image seen through each lens are roughly similar to a patch in the non-light field JPEG.... just skewed to a slight different place.
This isn’t 100%... new image can be exposed by looking around a corner, shifting reflections, etc. But each lens requires adding less and less to your compression table.
So really, the light field is an index into a slightly expanded compression space more than it’s a whole new set of data.
As for the complexity of rendering... it’s just a JPEG with a gnarly projection equation. There’s nothing fundamentally more complicated about that... although it’s complex math.
As with anything it takes time for code to mature, but in the limit there’s nothing fundamentally more complicated about it than playing compressed video.
There’s probably some efficiency to be gained with custom compression algorithms and custom hardware, although I don’t know enough about existing video compression hardware to say for sure how far the existing chips will get us.
11 comments
[ 1.6 ms ] story [ 38.8 ms ] threadIt's a better approximation than you can make with a normal camera, so its applications are even broader. Instead of a normal camera image which captures the position of focused light rays on a plane, this captures the position and direction of light rays as they move through a volume. Personally I would not call such a versatile instrument useless.
Point being, don’t judge this without knowing more.
Definitely very cool; cooler than it sounds. You get to experience a summation of all the subtle touches missing from VR today - like how a reflection on something changes as you rotate your head.
But like with all stuff VR, to "get it" you really have to experience it in VR. Especially how different this feels from 360 pics (stereoscopic or not) or photogrammetrie isn't obvious I think if you haven't seen it yourself.
This isn’t 100%... new image can be exposed by looking around a corner, shifting reflections, etc. But each lens requires adding less and less to your compression table.
So really, the light field is an index into a slightly expanded compression space more than it’s a whole new set of data.
As for the complexity of rendering... it’s just a JPEG with a gnarly projection equation. There’s nothing fundamentally more complicated about that... although it’s complex math.
As with anything it takes time for code to mature, but in the limit there’s nothing fundamentally more complicated about it than playing compressed video.
There’s probably some efficiency to be gained with custom compression algorithms and custom hardware, although I don’t know enough about existing video compression hardware to say for sure how far the existing chips will get us.
https://www.androidcentral.com/google-reportedly-buying-lytr...