Light field based camera or dslr? Which should I choose?

Question

Should I buy a lytro (light field) camera or a dslr one? I know that the light field based cameras are new. Do professional photographers use it? Or will they?

Asked by kadaj

Answer

In a decade, light-field cameras and computational photography are going to be the mainstream. As computer power gets faster, smaller, lighter, and more power-efficient, the advantages will exceed optics — maybe not completely in a decade, and probably not in every case in a hundred years, but there will be a crossing point.

But the current technology on the market — the Lytro — is essentially a toy camera serving as a tech preview. You'd buy it in order to tell your grandkids that you had the first camera of that type. It doesn't really do anything useful right now: the images it makes are postage-stamp sized, and the nifty tricks the camera offers only work when you upload the images to the company's special web site. You can't really do anything yourself.

They've promised an SDK but as of Spring 2012 that has not yet materialized. If you're a hacker (in the original sense), you might find that interesting and productive when available. But even then, it'll be more about playing with the technology than getting real photographic results.

That doesn't mean a DSLR is necessarily the only option for good results. You may also be interested in a small-sensor compact camera, or a large-sensor "mirrorless" camera, with or without interchangeable lenses.

Answered by mattdm

Check more discussion of this question.

Is there reference code for implementing light field photography?

Question

Many of us here at photo SE have read about light field photography and related cameras from Lytro and Raytrix.

A few of the advantages it allows are post-capture software focus, scene perspective changing and manipulation, and 3D video.

Are there any reference software implementations or available demo code of lightfield imaging for the above features and the related algorithms? (I read that FFT/IFFT is the tool used in obtaining software-focussed images, but I do not understand the full details of this.)

I've heard rumors that Nvidia Cuda has some thing for it, but I need more details if anyone has them.

Answer

I'm not sure about reference code; this is relatively new, and mostly what I can find are papers, not implementations with open code. A key paper is Fourier Slice Photography, by Ren Ng at Stanford University — now, not surprisingly, at Lytro. There's an abstract of the paper here, with a few nice pictures.

This doesn't give you something that you can take and just drop into place, but it gives you some technical, mathy details of how to go from the output of a plenoptic camera to a usable result. Of course, that's not all: for more, take a look at Ng's 200+-page PhD thesis on the topic.

You will probably also find the materials for MIT's MAS.531 / MAS.131 Computational Camera and Photography helpful. Thanks to the OpenCourseWare initiative, you can download and go through all the course materials yourself (including audio from the lectures).

Check more discussion of this question.

What are the basic workings of the Lytro light-field camera?

Question

lytro.com describes their new light field camera as being able to capture the entire light field, rather than just one plane of light, thereby allowing for a whole new set of post-processing possibilities, including focus and perspective adjustment.

What sort of sensor could "capture every beam of light in every direction at every point in time"? How would an essentially infinite amount information be encoded and manipulated? Would there be a lens up front in any traditional sense?

Here is the inventor's dissertation: http://www.lytro.com/renng-thesis.pdf

Can somebody boil this down for those of us who are familiar with traditional technologies?

Answer

The easy way to think about this is as follows:

Imagine that instead of one camera, you had a grid of 100 cameras in a 10x10 array. When you fire a shot, each of them shoots at the same time. They will each have a slightly different view of the thing that you are taking a picture of. There are some mathematical models you can use to sort of "reverse engineer" the image and rebuild it in different ways. That's what this is kinda all about, except that instead of 100 cameras, you have thousands, and they are all formed by an array of lenses just above the sensor plane. So the image that comes out of the camera sensor has a bunch of circles of partial images that each differ from the one next to them just slightly. Then they use math to reassemble a single image from those partial images.

Check more discussion of this question.