Hacker News Reader

Comment by rayiner | Hacker News Reader

> Refocusing after a picture has been taken is interesting, but useless unless your camera/phone/whatever doesn't focus properly.

Ever try and take a picture of a kid? My Nexus 5's auto-focus can't keep up with my toddler at all. Getting this technology down into a compact shooter or cell phone sized camera would be a huge leap.

View on HN

You will never get it in a camera the size of a cell-phone camera. This technology relies on the fact that each patch of their lens sees a different image of the subject; it's like stereo, but with a single lens, and many views rather than two. It fundamentally has to have a large-diameter lens.

You will never get it this small/big/fast/powerful/affordable is what has been said about a lot of industries/technologies in the past.

Question is, will Lytro be able to do it? Will they have the money&perseverance? With their financials they are not likely to move fast enough on this, 5mp equivalent in 2014 is not very impressive. I shelved my Lytro v1 after a couple of days because of the poor image quality, not because of the 'gimmick' disappointing me.

Pelican Imaging will be bringing out similar technology in cell phones this year or next. They're using a camera array rather than a single sensor + MLA.

What's the limitation on the size of each lens?

I am not sure what you mean by each lens, since there is only one lens with this technology. The diameter of the lens has to be comparable to the separation between the two lenses in a stereo system. For meaningful stereo effect at distances we would find interesting (say, a few feet between you and your toddler), this has to be maybe an inch; I don't see a smaller lens being very interesting.

For a typical cell phone, the hyperfocal distance - beyond which everything is in focus with the lens focused at infinity - is maybe 6 feet; you can't get light field information, at all, for anything further than that. And it will be only a tiny bit of information for closer subjects; you can't take shallow depth-of-focus photos with a cell phone, and you can't apply this technology for the same reason. BTW, motion blur is likely a bigger problem for cell phone photos than focus.

Almost all DSLRs and I assume many point-and shoots have an automatic focus tracking mode for moving subjects. Canon's is called AI servo. That said, most people don't move beyond the "auto" setting on their high-end cameras, so I can understand the quest to simplify.

Autofocus in smartphones will hopefully get better over time, but if your options are an overall superior DSLR or an equally large camera that just gives you some flexible (albeit one-dimensional) focusing benefits then I think the choice is clear. Practice a little more with your DSLR and wait to see if Lytro can be implanted into a better supporting cast of parts at a reasonable cost.

I have a Canon T3i and 2 young kids. AI servo is the right idea in theory, but on even a lower-end DSLR it just doesn't help much. If a kid is running towards the camera, it will keep refocusing as the subject moves, but it's always focusing on where the kid was a split second ago.

IIRC the higher-end cameras have algorithms to track the subject and predict where the subject WILL be and focus accordingly. However, I have many friends with Mk III's and they say it's a very hard problem that even the high-end DSLR's don't solve. As you can imagine it's not actually a solvable problem with the present AF latencies as the subject can change course after the algorithm guesses. And when you're shooting shallow DoF even a small mistake will kill the focus on the shot.

The question is whether the Lytro can operate at 1/200 or faster to solve that problem. The specs say 1/4000 is max shutter speed, but IIRC the actual capture speed isn't that fast...

Reply on news.ycombinator.com