A camera that never overexposes

Wouldn't this be neat..

http://petapixel.com …-overexpose-a-photograph/

Hrm,

Interesting.

Very best,

That's interesting.

From the article: “No more will photographers or even ordinary people have to fumble with aperture size and exposure length,” writes lead scientist Hang Zhao. “The algorithm would enable people simply to click the camera button and let the computer deal with exposure problems.”

What happens, I wonder, when we want to "overexpose"? And how can we make silky waterfall photos unless we fumble with exposure length? How will we control depth of field unless we fumble with aperture size? Hmmmmmm.

It sounds almost as if the technology is aimed at turning photographers, who know how to and want to control their cameras, into "ordinary people" who don't know how and don't want to know.

It wasn't all that long ago that AE & AF were pipe-dreams. Remember turning a lens-mounted aperture ring, or advancing your own film? I can't help but wonder how many of today's photographers would struggle with an all manual film camera. Almost forty years ago, I shot semi-pro soccer with such a beast. I think we tend to forget how far we've already come.

I could see this being a very useful feature and would welcome it in a heart beat. Just because the camera supports this type of technology doesn't mean that all manual control would be taken away just as a camera with AF and auto metering does not take away the ability to set things manually. But doing away with bracketing and blending HDRs, no longer need GNDs, that's a huge feature.

I guess in that situation you would have to process each individual situation in order to set both the white and black point in the final image. Really though all it seems to be is a useful way to extend the upper limit of exposure. In other words an extreme form of ETTR, where you can expose for the deepest shadows, without blowing highlights. Of course it will require us to record the data at much higher bit depths too.

Alan

Tedder wrote in post #17675807
That's interesting.

From the article: “No more will photographers or even ordinary people have to fumble with aperture size and exposure length,” writes lead scientist Hang Zhao. “The algorithm would enable people simply to click the camera button and let the computer deal with exposure problems.”

What happens, I wonder, when we want to "overexpose"? And how can we make silky waterfall photos unless we fumble with exposure length? How will we control depth of field unless we fumble with aperture size? Hmmmmmm.

It sounds almost as if the technology is aimed at turning photographers, who know how to and want to control their cameras, into "ordinary people" who don't know how and don't want to know.

I haven't had time to read their actual paper yet, but it sounds like they're using a multi-count sensor site, rather than a single-read sensor site that is the current norm. They are talking about a new (and better) way to read a digital sensor, not an AI that is picking settings for you.

If you are recording to jpeg, then yes the computer system is going to be harder for you to take creative control, but with a raw data file you can potentially do a whole lot more. The idea isn't that you point your camera at something, and the computer calculates your settings, but rather you pick your settings, and the sensor reads the light values no matter how over exposed they were going to be. So you could, in theory, expose for the back wall of a long straight cave in a white cliff side in full sun, and you still have readable highlight data on the cliff face rather than a pure blown out white.

Depending on their read method and data formatting you can potentially even do neat things like adjusting over time, and reviewing your photo almost like a short movie clip. In bright light you could capture a 30 second exposure, and then pull a 1/8000th of a second exposure out of the front of it if the equipment is designed to output time and count values.


The problem with multi-count sensors has always been that they tend to be extremely noisy and kind of unreliable, and the handful from research projects I've ever seen have all been lagging way behind mainstream sensors in resolution by more than a few generations. The biggest problem they faced was clearing the charge build up in the sensor site mid exposure. While clearing the charge that site couldn't be collecting any more charge, and it took a decent chunk of time (relative to the high speed nature of things) to read and clear, so you would be losing accuracy. There is also the issue of how you're clearing them: Do you poll each one in turn in an orderly fashion? Do you design each site to self trigger when full and dump to some kind of buffer? It can be easy to end up adding a LOT of extra circuitry into the sensor to handle this kind of thing, which then often ends up greatly reducing the total area of your sensor sites.

I'd love to see new technology that increases dynamic-range capabilities, so I'm not really poo-pooing what these folks at MIT have come up with. I just thought the quotation was kind of awkward and funny.

I'm not going to pretend that I follow your comments about "read method" and "data formatting," but the possibilities are fun to think about.

I can see this being very useful... we always want more bits for the RAW files, 12 versus 8, 14 versus 12. Now imagine 256bit or greater if needed

Quick, somebody head over to the Magic Lantern forums and post this link up there!

In all seriousness though, ML already has a feature somewhat similar to this with DualISO, and I *believe* it can autodetect the necessary shadow/recovery ISO.

Sounds a lot like my therorized "per pixel auto ISO adjustment", made simpler.


Of course, we do have highlight tone priority now, but that's far less flexible.

Will probably see implementation in military, surveillance cameras and medical imaging before it makes it into photographer/consumer hands.

I can already envision the newb photog questions on POTN.

"When I looked at the scene the sky was lighter and the sun was making the water white. Is there anyway I can fix this now?"

Unfortunately, the technology they show is spent making a sensor with terrible dynamic range appear somewhat decent, looking at the kind of shots they made, I could have repeated all of them with my 645Z in a single shot with better results.

By the time this starts becoming practical, 15+ stops DR will be the norm for your typical professional camera, and there honestly aren't that many scenes which could make use of more.

Admittedly this may be cool for cameras used in science more than anything.

Kolor-Pikker wrote in post #17678370
Unfortunately, the technology they show is spent making a sensor with terrible dynamic range appear somewhat decent, looking at the kind of shots they made, I could have repeated all of them with my 645Z in a single shot with better results.

By the time this starts becoming practical, 15+ stops DR will be the norm for your typical professional camera, and there honestly aren't that many scenes which could make use of more.

Admittedly this may be cool for cameras used in science more than anything.

True but I do regularly shoot at a location that can require around 30 stops of DR. To be honest most of it is a thin layer of back lit cloud, and I really don't mind it blowing out. My real issue is that exposing for the subject infront of that cloud just causes the sensor to swamp the edges because the contast ratio is too high. At the moment there seems to be no answer to this problem.

Alan