Interesting innovative camera

[simsurace]

I found this talk interesting, both for the innovative way to look at camera design and the clarity of explanations. It will be interesting to hear people's ideas about it! It takes a bit of time to watch it all, but in my opinion is totally worth it.

https://vimeo.com/143690545

EDIT: I should add a few details. The talk is about the design of the Light camera (https://light.co/) and imaging, especially computational imaging in general.

[Frank Fremerey]

The limiting factors are

1) computing power of the postprocessor
2) data storage
3) battery capacity

Data storage is a problem when you try to store 30MP x 16 in RAW. That is a lot of Data and even a 64GB card will be filled pretty soon. During the processing you have to temporary store all of these data.

Postprocessors get better but a lot of computing power means a lot of battery usage and a small camera means not much space for a large battery.

My guess is that the first generation will be take a shot, then wait for a long time before you can take the next shot. A responsive camera will only be there after a few generations.

[simsurace]

The limiting factors are

1) computing power of the postprocessor
2) data storage
3) battery capacity

Data storage is a problem when you try to store 30MP x 16 in RAW. That is a lot of Data and even a 64GB card will be filled pretty soon. During the processing you have to temporary store all of these data.

Postprocessors get better but a lot of computing power means a lot of battery usage and a small camera means not much space for a large battery.

My guess is that the first generation will be take a shot, then wait for a long time before you can take the next shot. A responsive camera will only be there after a few generations.

Good points. In the video he does mention that the battery should last for 400 shots and there is an optional grip with more battery in it.
The internal storage is projected to be 128 GB at least. That does not remove the constraint of processing. He mentions that some things will be processed only after the camera connects to a computer. I guess you have to keep the raw data until then, but I suppose they have some smart tricks up their sleeves to compress the data somewhat as it is captured. It is highly redundant data after all, you want to keep some redundancy in order to compute pixel depths etc. but maybe you can throw away some as well. Plus camera phone processors have become pretty powerful lately.

[John Geerts]

I have seen the video. It's  a gadget.

Apart from the points Frank mentions.  It's about stitched photo's with different lenses, beyond any control and dependent on the program of the makers / camera.

It has a fixed aparture, and the problems with depth or field are not answered.  And looking at the size of the 'prototype'  it's not small too     

[Frank Fremerey]

I want to have one to try and I want to shoot it in RAW to postprocess BOKEH anmd DOF at home. The multi perspective shooting offers a lot more information than a conventional shot!


@John: The DOF/BOKEH management is explained but may be the guy does not explain it very well or the idea is quite unconventional:

A classic lens has a certain dimension and any point on the lens produces a picture of a slightly different perspective. In a clasical lens these pictures are then mixed together analougous. The mixing is done digitally in the "light" camera, so DOF and BOKEH can be controlled after the shot. In the classical lens you can only choose the Aperture as the only parameter.

[John Geerts]

Yes Frank it's mentioned, and explained, but   and that's my point, there is a kind of 'uncontrollable digital brew' to come to a resemblance of a kind of depth of field with the lens-blur you want. On top of that it's combined of pics with a number of lenses. 

With a classical lens (and at the desired aparture) this can be done in post-processing by the way, but then you control the buttons.
 

[simsurace]

I have seen the video. It's  a gadget.

Apart from the points Frank mentions.  It's about stitched photo's with different lenses, beyond any control and dependent on the program of the makers / camera.

It has a fixed aparture, and the problems with depth or field are not answered.  And looking at the size of the 'prototype'  it's not small too   

Computational imaging has been around for quite some time, but not too long ago, doing the sort of computations required in this camera would require a massive computer and lots of time. Bringing these technologies to a commercial portable device is the main challenge they are facing. We will see next fall how it pans out.

The lack of control in stitching is not a minus, but a plus, you don't want to do this by hand trust me. Instead, you get control at a higher level over things like dof, bokeh characteristics and focus distance after the picture has been taken, as Frank and the co-founder in the video nicely explained.

When stitching two shots which have parallax (maybe you have tried doing this), you can make only objects in one plane match. This is your virtual focus plane. Objects in front or behind that plane will have a double image. By mixing the two versions in a smart way, you can control the amount and characteristic of blur. Having more than two shots will make it smoother. A similar computation is done by an ordinary lens, but it is done in an analog way by combining the electromagnetic waves that make up the light, whereas this camera first digitizes the waves and then combines the resulting data. Phase information is retrieved from the data thanks to the different viewpoints (they can reconstruct a depth map, but don't need a fancy sensor as the LightField camera does, only ordinary cellphone cameras).

[simsurace]

Yes Frank it's mentioned, and explained, but   and that's my point, there is a kind of 'uncontrollable digital brew' to come to a resemblance of a kind of depth of field with the lens-blur you want. On top of that it's combined of pics with a number of lenses. 

With a conventional lens, the brew consists in having to accept a given lens design, which does all the computation on the light waves. Most of us don't design our own lenses, so we don't control that (apart from picking a different lens, but that's only a couple of choices in a problem with thousands of variables).

One additional point which is important is that doing the lens computations in the digital domain gives one the flexibility of getting rid of all aberrations, allowing the imaging process to be diffraction-limited. This is truly a step forward for such a small device. To do this in the analog way often requires very expensive glass, lots of aspherics and it adds up to size and weight. Aberrations are sometimes wanted, but you can control even that (if they allow for it in their software, but they are open to suggestions and I'm going to suggest that), so if you want to mix in a little spherical or chromatic aberration, you can have it.

[Frank Fremerey]

Yes, we are in a very speculative domain here.

The concept is amazing but the challenges faced by the designers are abundant.

When they make this work we will see a completely new kind of camera.

It feels very much like the early announcement and preorder phase of the first Tesla car many years ago. Currently they build their third low volume model after roadster that is no more produced and the Model S which is their only product at this moment, the Model X, and in the background preparing the first mass production car the Model 3. Tesla is still losing money, so the question is: Will they survive until they live up to their promise? I very much hope so because Tesla is currently the only option for cars significantly better to the environment.

In that sense: will "light" survive until they can live up to their promise?

[simsurace]

This is always hard to predict. If they manage to launch it in a phone which is bundled in contracts of a major American cellphone company, it might take off.

Currently, the asking price for pre-orders is still fairly high (1700$) if all you expect is slightly-above-cellphone image quality. If on the other hand, the image quality even remotely approaches that of large-sensor cameras with very good glass (given my understanding of the technology, this is a high, but not an entirely unreasonable expectation), this would be a very attractive price for such a capable imaging device in a small form factor.

For me, the biggest photographic drawback is that the FOV is limited to 35mm-equivalent. I hope they will do something to have wider angles as well.

[Frank Fremerey]

If you take several 35mm shots you can stitch them.

Given the builtin post production tech this should be no problem.

Joke: how to define the entrance pupil of a 10 cam virtual telescope?

which reminds me of some huge Telescope in Chile that consists of
4 smaller telescopes and emulates a huge telescope in software...

[simsurace]

If you take several 35mm shots you can stitch them.

Sure, I thought of that
Works well for static stuff, but for moving subjects it's going to be trickier.

[simsurace]

which reminds me of some huge Telescope in Chile that consists of
4 smaller telescopes and emulates a huge telescope in software...

Great observation.

[Frank Fremerey]

http://m.phys.org/news/2015-04-virtual-telescope-black-holes.html

[Frank Fremerey]

Not only thew US of Americans can now bet on a race horse that is not even born....

Good news! The Light L16 Camera will be available for pre-order in the UK, Germany, Italy, Netherlands, and France starting November 7th. Act now to reserve your camera for international shipping.

    Stunning DSLR-quality images up to 52 megapixels
    35mm-150mm true optical zoom
    3 Fast prime lenses
    Onboard editing and wifi

For more details about international supply & shipping information, please visit spot.light.co/international