World's First Full HDR Video System Unveiled

Zothecula writes "Anyone who regularly uses a video camera will know that the devices do not see the world the way we do. The human visual system can perceive a scene that contains both bright highlights and dark shadows, yet is able to process that information in such a way that it can simultaneously expose for both lighting extremes – up to a point, at least. Video cameras, however, have just one f-stop to work with at any one time, and so must make compromises. Now, however, researchers from the UK's University of Warwick claim to have the solution to such problems, in the form of the world's first full High Dynamic Range (HDR) video system."

Comments

Wow!

[snspdaarf]

Now p0rn can be filmed in sunlight and shadow at the same time!

More interesting than 3D?

[Anonymous Coward]

Personally, I think the HDR screen described, with HDR videos, would be more interesting and immediately useful than the ever-so-commonly-advertised but ever-so-rarely-purchased "3D" screens.

Re:More interesting than 3D?

[penguinchris]

Yes, but... that's not what it actually looks like to our eyes. Same with HDR still photography, which I've dabbled in a bit. The actual video or photograph may be HDR, but in order to display properly on an inexpensive screen, the depth is compressed. Thus, you need expensive screens to properly see. That's not to say you can't get impressive results that are indistinguishable from the real thing to the untrained eye, but the ability to truly display HDR video has the potential to blow people away - it'd be able to trick our minds into thinking it's a window into another world, even without 3D... and combined (assuming the 3D is done well) it'd be astonishing.

Not so fast there son

[Anonymous Coward]

Didn't a pair of guys do this last year using a pair of DSLR's and a beam splitter?

Also ,unless someone is building a HDR display this is all pretty academic, HDR images have to have their range compressed and then tone mapped in order to be displayed via conventional means, this is normally terribly unsubtle and results in an image that looks not entirely unlike it was rendered using 3d modelling. If we are going to see another big shift in display (read: TV) technology in the next decade I would much rather we moved away from the sRGB / YUV colour space than started fucking about with HDR content, what's the point of trying to take advantage of our eyes exposure latitude if we can only render 1/3 of colours?

Re:Not so fast there son

[UmbraDei]

Correct, it's been done before using 2 Canon 5D Mark II's, as seen in this clip: Soviet Montage Blog

Re:Not so fast there son

[MightyMait]

If you read the article, the author mentions both the folks in SF who did HDR video previously as well as the fact that the Warwick team have indeed developed a new HDR display for their system.

Also mentioned is that the new system generates 42GB/minute of data to capture images with 20 exposures per frame. My backup nightmare just grew large fangs!

Re:Not so fast there son

[zeroeth]

Speaking of 3D rendering, most of them output HDR which would be awesome to see without being tone mapped!. (LuxRender has built-in Reinhard tone mapping) And since we are on the topic of HDR.. this is what it is NOT http://www.cs.bris.ac.uk/~reinhard/tm_comp/flickr_hdr/The%20Problem.html (Reinhard discusses the blown out tone mapping heavily prominent on flickr)

Cool stuff

[Monkeedude1212]

I first learned about HDR from Valve, during one of the developer commentaries on one game or another... (Lost coast maybe? Anyways) They were trying to explain how Bloom is done in video games, and certain other effects like how walking out of a dark tunnel to bright light will affect your vision for a tiny bit, as your eyes need to adjust to the new lighting conditions.

Thats when I started looking it up and yeah, basically the idea is that you take one shot that is under exposed (dark), one shot that is over exposed (light) and one that is properly exposed, and as many more in between as you want. Then you feed it all into a bit of software which takes the richest colours and lighting conditions from each photo and imposes them into one single image, so the dark corners remain dark and the bright lights remain bright and the vivid colours are still vivid. Its quite cool stuff.

I'm a little curious as to how this is working, is it managing to encode the HDR real time into it's range compressed and tone mapped beauty at least 24 frames per second, or does it merely record the 3 or more images simultaneously and then take a few minutes afterwards to do the encoding? The first I think would be more impressive, but not really necessary.

Re:Cool stuff

[Doogie5526]

I find it kind of funny that HDR means the opposite thing in photography versus video games
http://img194.imageshack.us/img194/7391/1244894383293.jpg (pulled from some old digg post)

Traditionally games render the world and keep it between 0 and 1 (zero being black/completely dark and 1 being white). HDR is computing values above and below and clipping so things that are blown out (like reflections and highlights) are super white. I think it was an update to Half Life 2 that first did this in a commercial game.

In photography, they take multiple exposures and stick them in to an HDR image. Then, they use tone mapping to convert it to an 8-bit visible image. Tone-mapped images are generally called HDR, even though that's a misnomer.

Re:Cool stuff

[nomel]

Unless you have an HDR screen, this would require an automatic tone mapping. The thing about automatic tone mapping is that you have to decide what intensity information to throw out since you only have 256 values that you can display. For instance, using a 14 bits per color channel canon DSLR sensor, if you want to look at the image on your screen, this means you'll have to thrown out 98.4% of your intensity values. It is extremely important which values you decide to throw out, especially considering there's usually a subject or subjects in a photo that you want to keep visible.

By the way, this 14bits gets you about +/- 2 stops...the camera they're talking about gives you 20 stops...that's an *incredible* amount of intensity information (giving the file size). Really this is more of a solution for filming a scene once and not having to worry about if you camera exposure is set correctly, which *is* extremely valuable.

Now, viewing HDR movies? Not in theaters with any sort of current projection technology with reasonable ticket prices. The projection bulbs would have to go up probably 20 times in brightness, keeping similar crappy projection theater black levels. And, how do you deal with the ambient light coming off of your now incredibly bright white screen and bouncing off of the audience? At home, do you really want a tv that bright? From this bit-tech review, "The light from the box was so bright, or indeed, was of such great contrast with the surrounding area, that it almost hurt to look at.".

Re:Cool stuff

[The13thSin]

I would like to correct a mistake prevalent here and in the news summary: common camera's do NOT get 1 (or 2) stops of light information (the difference between black and white). In fact, camera's like the Canon 7D have about 11 stops of dynamic range [source] and professional video camera's like the Red One have about 13 1/2 stops of difference between black and white [source]. Still, as X stops means 2^X times the light difference, going from 13 1/2 to 20 stops is a pretty huge deal.

Another misconception: the amount of bits per channel only indicates precision, not dynamic range. Of course, when the researchers in the article created a 20 stops camera, they needed much better precision to get similar quality in the same range as the current camera's, which leads to the quoted 42 GB per minute uncompressed video stream.

(Please note: DSLR camera's like the Canon 7D can detect and save more dynamic range than is apparent from the JPG's they create and the extra information is saved in the RAW file, which allows you to change exposure settings at least 1 stop in post processing without (noticeable) drop in quality.)

So how about some decent framerate?

[NeutronCowboy]

I mean, we have 1080p 3D stereovision with full-micron surround color effects, and yet, movies still stutter like mad on a fast pan because that damn 24 fps capture rate just can't keep up. Is it really so much harder to capture 60 fps and encode than it is to do a working 3D effect? I'd pay more for movies that have reliable framerates in the 60 Hz range than I would for 3D.

Re:So how about some decent framerate?

[Doogie5526]

Roger Ebert asked the same thing (on page 4)
http://www.newsweek.com/2010/04/30/why-i-hate-3-d-and-you-should-too.html

I think there's a couple reasons. The first, and probably most significant, is nostalgia by film makers. They love the motion blur of 24fps. It helps evoke the "feeling" of film. Every film student I know either wants to shoot or convert their footage to 24fps. There is a noticeable difference. When you start increasing the resolution and frame rate, you lose motion blur and it starts to look like home video or video games (when generally don't compute motion blur at all).

Another big issue is the amount of light. When you have more frames in a second, each frame has less light to suck up. It's a big issue with high-speed film. Having sensors that are more light-sensitive is a fairly recent thing (combined with advanced noise reduction) and will continuously get better.

The stuttering is something cinematographers keep in mind when shooting (or at least, they should). I read an article about shooting imax and they said the biggest problem was the stuttering. They're also using 24fps, but the screens are much larger. When you pan, the object could jump 2 to 3 feet per frame. They intentionally had slower pans to compensate. You noticing this is probably a side-effect of larger theatrical screen and larger tvs at home.

Re:So how about some decent framerate?

[ChrisMaple]

The same motion detection that video compression uses to reduce data rates could be used to interpolate as many frames as you wish between two existing frames. Goodbye stutter. As software projects go, it's not even a particularly difficult job.

Re:So how about some decent framerate?

[nomel]

It's because it's not pleasing to the eye. 60fps movies look very strange...like home videos. The 24 fps is what gives them that "movie look". If you look at some example vids from some of the newer consumer cameras that can do 24 and 60fps...you'll see the huge difference it makes.

Re:So how about some decent framerate?

[futuresheep]

24fps has nothing to do with being pleasing to the eye. 16fps was common before this, Edison was pushing for 48fps. 24fps was a compromise that was chosen simply because it was the slowest, meaning cheapest, speed that allowed for good sound quality. http://en.wikipedia.org/wiki/Film#Technology

Modern cameras have _much_ more than one f-stop

[aacosta]

Video cameras, however, have just one f-stop to work with at any one time, and so must make compromises

Just to name one example, the Red One has a wonderful dynamic range of 11.3 stops. http://en.wikipedia.org/wiki/Red_Digital_Cinema_Camera_Company

Re:Modern cameras have _much_ more than one f-stop

[muridae]

The summary is just plain wrong, and the article may be as well. First, there seems to be some massive confusion between f-stops and dynamic range 'stops'. An f-stop is your aperture setting, and is part of the control that determines how much light gets into the camera. If I go out and desire to take an HDR picture of something, the f-stop is the last control I will use in setting each exposure. The f-stop has the side effect of changing the depth of focus, thats covered in photography 101. If you change that in a set of pictures, some things will be in focus in one frame, while out of focus in others. It doesn't look that nice once post processed.

On the other hand, a dynamic range stop is just notation for double the amount of light. If someone said "That film has about 9 stops of resolution" you would know that physically the brightest area on a picture would have 2^9 times as much photonic flux. Or you would be more camera focused, and know that the film would only record detail in the 4.5 stops above, and 4.5 stops below what ever you set the exposure for. An object 5 stops brighter than what you were focused at would be a washed out blur, and something 5 stops darker would be total shadow. A quick run through google suggests that Kodachrome, the legendary film, could record only about 8 stops dynamic range. The human eye can pick up something closer to 24 stops. GP's Red camera records 11.3 stops. Some people will claim that a digital camera gets as many stops as bytes, but that is only with the analog to digital conversion is logarithmic, and so is the display it is shown on. Mine runs about 7 stops, depending on other settings.

So, what's that got to do with this camera? I suspect what the article meant to say is that the camera captures 20 stops of data at 30fps. Better than the Red, better than almost any film in existence. It is doing the same thing in a single shot that other cameras do in several. All that will mean is less blur in HDR video, since subjects won't move irregularly between exposures. One would still have to tone-map the output down to a range that it can be displayed for printing, projection, or dvd.

This is not about aperture, nor about ISO

[RichiH]

It does not matter that a camera can only have one aperture and one ISO setting. Our eyes have only one iris, as well. What matters is that our retinas & brain have a dynamic range that trumps CMOS/CCD sensors. Oh, and the fact that our eye cheats by seeing more colour in the middle of the retina and more bright/dark & movement in the corner of our eyes.

That being said, I am looking forward to anything that extends the dynamic range in both cameras _and_ displays.

Re:As a hobby photographer...

[moonbender]

Well, there are situations where you need multiple exposure to avoid clipping some details. E.g. images like this one: http://hdrcreme.com/photos/23496-Entrance-of-a-Small-Church There is no way a consumer camera can get the detail of the interior of the dark room (and the colorful window) as well as the bright outsides.

HDR motion picture cameras are already shipping

[aibrahim]

RED has shipped its first two production EPIC-M cameras which has a feature called HDRx, which allows up to 18 stops of DR in a single exposure for every motion picture frame. It doesn't require a beam splitter or any other gadgetry.

Peter Jackson has a number of them he's using for the Hobbit. I think the latest Spiderman is shooting with it too.

It does that at 5K, which is 5120x2880 resolution.

As to comments that HDR is better than 3D, or that you don't need lighting... they are unfounded. You still need lighting to create the precise mood you want. The advantage is that you can now create that mood more easily in more lighting conditions. This is especially important in conditions that the film maker can not control. The first RED demo was a shot from inside a barn out the barndoor into the Arizona desert. The camera held detail in the shadows inside the barn and in the sky and on white surfaces in direct sunlight.

The normal solution to that lighting situation is to pour about a hundred thousand watts of lighting into the inside of the barn, hope nothing catches on fire and that you are close enough to the sun

The First?

[CoolGuySteve]

Didn't Autodesk Toxik 2008 already do HDR compositing with RED cameras?

Not only that, didn't they sell it to real live customers?

This is not the first, it's not even notable, frankly

Re:f-stops are bunk

[dfghjk]

You defined a stop, not an f-stop. You are not alone in the confusion. There is a big difference between the dynamic range of the sensor, in stops, and the exposure settings necessary to utilize it. The author of the article doesn't understand that either.

A lot of confusion about HDR.

[Rothron the Wise]

HDR photos you find on the web are actually tone mapped photos. They were HDR when they were captured, or when different exposures were combined into a single image, but after that stage they were tone mapped in order to make all the details visible on a conventional display.

Tone mapping is something we may stop doing when we have proper HDR displays like in this article. A display like that will more closely resemble the real world, and tone mapping will be unnecessary because our eyes can handle high dynamic range images just fine.

The perfect HDR video system would be one where you could film inside of a dark cave and you would see everything on the screen after your eyes had adjusted to the dark, and when the camera moved outside into the sun the intense brightness of the screen would make you squint.

Cheesy tone mapped HDR photos make your eyes hurt for totally different reasons.