Slashdot Mirror
The Lytro Camera: Impressive Technology and Some Big Drawbacks
waderoush writes "The venture backers behind Lytro, the Silicon Valley startup that just released its new light field camera, say the device will upend consumer photography the way the iPhone upended the mobile business. This review takes that assertion at face value, enumerating the features that made the iPhone an overnight success and asking whether the Lytro camera and its refocusable 'living pictures' offer consumers an equivalent set of advantages. The verdict: not yet. But while the first Lytro model may not an overnight success, light field cameras and refocusable images are just the first taste of a revolution in computational photography that's going to change the way consumers think about pictures."
Right now, it seems like the majority of Lytro pictures are technology demos, a fire hydrant in the foreground and a building in the background, or some equivalent, which just invites you to click both and move on. You can just hear the enthusiastic early adopter in the background of these pictures saying "OK, _now_ click the building! Whoa! Cool, huh?!". These shots are, to my mind, the photographic equivalent of arrows or spears coming out towards the audience in early 3D movies. Gimmicks which break the fourth wall, saying "Hey, remember, you're looking at a Lytro (tm) image, not just anything!".
I can't wait for real photographers and artists to actually find situations, styles and aesthetics where Lytro sorts of cameras can be used in a way that both effectively uses the new capabilities of the format _and_ produces something artistically and aesthetically wonderful. I think the technology has a ways to go, but right now, the biggest problem facing Lytro (and light field photography) is that it's a new medium that nobody has a clue how to use effectively.
Until we reach that point where people see a great Lytro picture and actually feel inspired, it's going to be tough to sell what is currently a low-spec camera with one big gimmick. So, if you want Lytro to take off, buy one for the craziest artist you know.
The capabilities of light field cameras have that fun 'technology indistinguishable from magic' touch to them that the impressive-but-evolutionary spec bumps of markedly superior conventional digital cameras don't(It's like playing with your favorite eccentric retro computer from before the Great Standardization: at this point, anything that old is a painfully limited toy; but it is different. Your top-of-the-line-screaming-monster of a PC, on the other hand, is brutally capable and impressively cheap; but practically point-for-point familiar to the p90 running Windows95, with all the performance related numbers bumped by a few decimal places).
Unfortunately, though, the move to release it at a (barely) 'consumer toy' price point really led to a product slightly too compromised to be useful: The optics you need for the light field capture eat so much of the sensor's available resolution that the resolution of the images you can get out of the thing is hovering slightly below 1 megapixel. Yes, the ability to spit out that paltry image at all sorts of focuses, after the fact, is damn cool; but for $500, you could get a high end P&S that could iterate through a series of 10MP shots at different focus points, at time of shooting in a few seconds, netting much of the benefit along with resolutions that wouldn't be ashamed to show up on a $20 webcam.
I'd love to see the same technology applied at a price point and form factor where the sheer sacrifice of available pixels wouldn't be so keenly felt.
Let's list some of the significant drawbacks of this first version which we can realistically chalk up as a technology demo:
* Camera is shaped weird and appears awkward to use. If form follows function, I'm not sure what the function is.
* Cheap last-gen LCD display.
* Output is only 1MP (1024x1024).
* Sensor is really small
* Lens is cheap
* Limited depth of field
* Raw light fields have to be sent to Lytro server for processing
* Only a handful of focus points can be chosen
* In focus range is limited
* Photos are converted into lame Flash animations
Now, let's re-imagine this as a serious photographers tool a few years down the road:
* It's a DSLR with real interchanegable lenses and huge hi-rez LCD display
* Let's say the camera can even magically switch from "classic" to light field mode with a toggle switch.
* Huge full frame sensor allowing light field output at 6+MP with high dynamic range and low noise at high ISOs
* Depth of field choices much broader and limited only by lens chosen
* Effective focus range is much improved
* Raw lightfield processing can be done on your local computer, allowing precise control over number and position of focus layers. Alternately, assuming processing speed is available, perhaps focusing points can be chosen in real-time within the finished image blob.
* Output as multiple jpegs, flash or HTML5, etc.
Now what?
Well, you still have these limitations if you use light fields:
* You're basically giving up some amount of image resolution for the ability to focus after the fact. DSLRs and even consumer cameras already have excellent auto-focus modes that when used properly generally nail focus in decent light. It's not the biggest or even second biggest problem I see in photos online. Bad composition and inadequate lighting are generally much bigger problems.
* If you chose the wrong focus point when shooting, sure you can fix your mistake, but if focus is off due to camera shake or motion blur, you're SOL.
* It's basically useless in images with large depths of field (think large landscapes where everything is essentially in focus)
* Makes no difference on a printed page, except you have one more tweak available during editing.
* Still gimmicky. After everyone has played around with a few of these photos interactively, they're bored and move on.
It's a single photosensor. The lens array and maths are doing the hard work. Therefore, although the data processing requirements may be very data intensive, the actual image should be the same, or very close to the same, as an image taken without the lens array. The maths should be implementable fully in hardware such that all processing can be done on camera at video speeds, so there is no reason that this couldn't be done. The issue would be making a cohesive focal point between frames. Having to focus a film frame-by-frame would take a lot of time and would be something only film studios might be willing to do, but would be too annoying for consumers.