Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Lytro Unlock – Making a bad camera slightly better

github.com

Post reply on HN
Print view View original

31–40 of 74 posts

Re: Lytro Unlock – Making a bad camera slightly better

#31

Post by matsemann »

> Killer feature of this new technology was the ability to refocus the image after it was taken! (...), the camera was trying to solve a problem that didn’t exist

A related camera technology, but which I've found great to use, is 360 cameras. The ability to re-frame (not focus) the image as I want in post is so great. No longer do I have to spend many attempts positioning a camera perfect, just shoot and later "point" the camera correctly. For instance when filming ski videos, it could take multiple attempts getting the angle correct to get exactly the person+jump in view. Now it just works.

Re: Lytro Unlock – Making a bad camera slightly better

#32

Post by ddalex »

weinzierl wrote: post #22
d1sxeyes wrote: post #14

Earlier quoted context omitted.

That’s a really interesting way of phrasing it I’d never fully considered: of course you are right, a physical lens is just performing predictable mathematical transformations on the light passing through it. Thank you, that was something of a revelation in understanding how these cameras actually work.

"just performing predictable mathematical transformations" Just linear algebra, to be precise. A lens is just a matrix and rays are simply vectors. Combining lenses is merely matrix multiplication and complex optical systems can be represented by the result, which is a plain simple matrix again. The Eigenvalues tell you interesting properties of your optical system. I've always enjoyed this part of physics because it…

Going on a metaphysical tangent, it is a bit weird that A LOT of physical processes can be modelled with linear algebra, and don't require something way more advanced ...

Re: Lytro Unlock – Making a bad camera slightly better

#33

Post by fhars »

weinzierl wrote: post #22
d1sxeyes wrote: post #14

Earlier quoted context omitted.

That’s a really interesting way of phrasing it I’d never fully considered: of course you are right, a physical lens is just performing predictable mathematical transformations on the light passing through it. Thank you, that was something of a revelation in understanding how these cameras actually work.

"just performing predictable mathematical transformations" Just linear algebra, to be precise. A lens is just a matrix and rays are simply vectors. Combining lenses is merely matrix multiplication and complex optical systems can be represented by the result, which is a plain simple matrix again. The Eigenvalues tell you interesting properties of your optical system. I've always enjoyed this part of physics because it…

Well, sort of... there goes quite a bit of materials science and engineering into making and shaping the glass of a lens to make it behave like a matrix (i.e. to extend the range in which this linear approximation to the actual behavior of a blob of glass is valid).

Re: Lytro Unlock – Making a bad camera slightly better

#35

Post by fhars »

ddalex wrote: post #32
weinzierl wrote: post #22

Earlier quoted context omitted.

"just performing predictable mathematical transformations" Just linear algebra, to be precise. A lens is just a matrix and rays are simply vectors. Combining lenses is merely matrix multiplication and complex optical systems can be represented by the result, which is a plain simple matrix again. The Eigenvalues tell you interesting properties of your optical system. I've always enjoyed this part of physics because it…

Going on a metaphysical tangent, it is a bit weird that A LOT of physical processes can be modelled with linear algebra, and don't require something way more advanced ...

That is actually deliberate, we tend to organize everyday life and the devices we build around processes that are easy to model, that is why many things look like linear processes and harmonic oscillators (the first two term of the Taylor expansion of the actual behavior). We change the type of spring we use if the current one starts to wear out early under normal operational conditions.

Re: Lytro Unlock – Making a bad camera slightly better

#36

Post by Zetobal »

worldmerge wrote: post #12

Very cool! Would it be possible to use this for NeRFs? I’m very new to them but it seems like they use light data heavily, at least it seems Luma AI does.

It wouldn't make sense to use these for nerfs because of the variable focus... and the low resolution. The depth prediction used by NERF is also miles ahead of the depth maps produced by lytro cams.

Re: Lytro Unlock – Making a bad camera slightly better

#37

Post by Frost1x »

daniel_reetz wrote: post #30

Author, if you're reading, a really cool thing would be to create a small array of these cameras (4 of them) side by side. The very wide baseline should allow you to enhance/deepen the refocusing effect and create a wide format image. I've always wanted to do this.

You can also extract depth information from each cameras lens array. In theory, if you combine that capability on a per camera basis with photogrametric and/or stereo imaging techniques, you should be able to get a fairly refined depth map of a scene passively if the scene is static or you have well synchronized shots. Something I also wanted to explore but never had time to do.

At the time, I always considered passive ranging to be a potentially more valuable feature of lightfield cameras more so than refocusing images.

Re: Lytro Unlock – Making a bad camera slightly better

#39

Post by GeompMankle »

jbandela1 wrote: post #38

On a side note, do you think think type of light field photography could be packaged in a phone camera? Right now we have multiple lenses sticking out the back of the phone. Could we replace all that with just a single large light sensor?

If I correctly interpreted your question as ending in "a single large light[field] sensor], no. Light field cameras take a single field of view and overall entrance pupil and then, given that field of view, they trade resolution for being able to specifically note which part of the entrance pupil saw the object in that low resolution way. As a person who specializes in physical optics rather than computational optics, my extremely biased standpoint is that light field sensors can take one really great camera and make it into a bunch of fairly poor cameras collecting 3D depth information. If you value 3D depth information and the ability to add realistic depth of field blur to a photo more than the 2D resolution and sensitivity/noise properties of the photo, lightfield is for you.

The reason we have multiple lenses sticking out of a phone is to achieve similar high image pixel counts at high sensitivity/low noise across wildly different fields of view (say 108 deg, 69 deg, 28 deg) and wildly different per-pixel angular fields of view in a cost effective manner. A single imaging block capable of seeing a whole 108 deg field of view wide camera image with the per-pixel angular field of view of the 28 deg telephoto imager is generally infeasible in a cost effective way unless you are IMAX.

Re: Lytro Unlock – Making a bad camera slightly better

#40

Post by AstixAndBelix »

rixrax wrote: post #26

Nice to see old Lytro hardware brought back to life. When they were launching the company, I was so excited about the possibilities. Maybe naively, I was really rooting for being able to take a photo, and then in their software (and later in Lightroom) be able to slide/brush-in/out points of focus. Alas, all we got were these gimmicky square pictures that you could click to bring various parts to focus.

As always, a company makes a very interesting piece of hardware available but locks it down with proprietary software so when it dies almost nobody can keep using it effectively.

At least these project of reverse engineering the file format or even the firmware can bring new life to these quirky cameras

Post reply on HN