Slashdot Mirror
HDR Video a Reality
akaru writes "Using common DSLR cameras, some creative individuals have created an example of true HDR video. Instead of pseudo-HDR, they actually used multiple cameras and a beam splitter to record simultaneous video streams, and composited them together in post. Looks very intriguing."
HDR
Focus Stacking
Panoramic Stitching
All in the camera, all 1-button easy to use, and all at once.
Can anyone give a brief rundown on what HDR is? I know it stands for "high dynamic range", but as someone who knows nothing about photography, it means nothing to me. What it has to do with overexposure/underexposure (to which the video refers)? Why is it harder to do with video than still images?
I've been a long-time fan of HDR photography, and was just thinking about ways that HDR could be implementing in video camcorders as well. Personally I'd like to see a correctly-exposed stream mixed in with the other two, as is common in photography, but even without that the effect is pretty darn cool.
By the way, in case any camcorder manufacturers are watching, consider this idea: make a video camera with three (or more) times the required number of sensors for the resolution you want to record at. Set the logic in the device up to use three unique sets of sensors inside to pick up three different sets of images, at differing exposure settings. Then have them saved separately so that they can be integrated later on for various editing effects - or have a mode where they are integrated on-the-fly for easier use by non-professionals. I imagine it would be expensive to make such a complex sensor and camera, but it might be easier to manage than multiple cameras as the folks in the article did.
William George
I had my first foray into HDR still photography recently and I have to say I'm very very impressed with the results. Certain night-time scenes look absolutely stunning using 4-5 exposures. Here's some shots by a friend of a friend: http://roache7.deviantart.com/gallery/.
Spheron had an awesome single-sensor HDR video camera demo at SIGGRAPH this year. It records 20 stops of latitude, and after some processing for debayering and whatnot, you get an EXR sequence. I got to see it live, in person, and stand a few feet away from the camera. The guy running the demo even let me play with some footage in Nuke on the demo laptop. I'm confused about why a hacked up beamsplitter based system would be so noteworthy, when the single-sensor method will suffer less light loss thanks to the simpler optical path.
I'm sure the guys who did this project are proud of what they pulled off, and it's probably a neat hack, but I have to assume they are sort of operating in a vacuum if they think they have really invented something newsworthy.
With frankencamera you could do HDR and a lot more things in an "intelligent" camera with software. In fact the first implementation in a mass consumption device was in the N900, it takes several photos, regulates exposition and other parameters to make that photo in a more parametrizable way that the iphone could do. But not sure if that would be enough for HDR video, if needs that the input, in real time, have different something at hardware level. In that case maybe something like this 3D camera would be needed. And could give some meaning to such devices... not only shooting in 3d, but in HDR video.
So, HDR video would help make movies look like ... video games??? Is it just me or does that video (that parent linked to) look amazingly like a (post-HalfLife2) game? I guess this should be a fantastic clue for game programmers who usually try to go the other way ;). Lack of HDR = more "realistic" video? (where realistic is defined by what people are used to). Find an algorithm to intelligently degrade the dynamic range in a rendering and CGI becomes more photorealistic.
That's pretty much down to our mental training that a photograph is a realistic representation of lighting in a scene.
This is similar to the mental effect which makes high frame-rate 60-90fps video look "fake" and less true-to-life to us, who have been watching 25fps movies for decades, despite the opposite being true.
In truth, printed photographs are terrible representations of light and instead rely on our knowledge of the elements to trick our brain into viewing lit scenes in the context of previous experiences. Digital photographs, capable of being artificially lit are much better, but still not as good as real life.
However, the best true-to-life representations of digital photographs is SDR tone-mapped HDR images. Look at the lights around you - your eyes DO see those blooms around lights, etc. Years of looking at standard photographs has trained us to believe that they're a great representation of real life - when they're not. They're simply the best we've been able to generally do.
Besides, eventually HDR will be the norm, and this entire line of conversation will be moot. By that time, they will be "a normal photograph". In fact, HDR techniques have been practiced for a long time now - heavily since the 80s. Many of the "great" published photos of our times were taken with multiple exposure techniques - we might just not realize it because we only see the final result.
"The true measure of a person is how they act when they know they won't get caught." - DSRilk
It's the other way around.
Even though we call it high dynamic range in videos and photographs, it's actually just compressing all the extra information from multiple exposures into a LOWER dynamic range, so we can manipulate/display it on our 8-bit screens.
Games, however - such as the Source engine after it got the HDR update with Half-Life 2: Lost Coast and Day of Defeat: Source, actually do increase the dynamic range of a scene beyond what your monitor can display. They underexpose and overexpose parts of the scene when transitions between light and dark places occur, just as your eyes would before they adjusted to the new light, or as a video camera would depending on what exposure the videographer chose. This makes it look more realistic - just take a look at a bright outdoor scene in Half-Life 2: Episode Two and check out how shiny objects in the sunlight have blown-out highlights that gleam brilliantly, and then look at the same scene in the original Half-Life 2, where that object would look flatly-lit and fake. The "non-HDR" looks more fake because the dynamic range is compressed so you can see all the detail everywhere, which also gives it that flat "game" look.
Of course, that last part is just my opinion - but I believe that in order to look more realistic, CGI needs to simulate the behavior of traditional cameras with a lower dynamic range (or that of your eyes before they've adjusted properly to bright/dim light). The everything-is-exposed-properly, compressed-dynamic-range look just appears fake to me, even though my eyes could probably perceive that range at the actual scene. I'm not sure why.
WHO NEEDS SHIFT WHEN YOU HAVE CAPSLOCK/ DAMN1
For the third of the Fast and Furious movies, we had to film at night in the spectacular Shibuya Square in Tokyo, with its many animated billboards and video screens. I really wanted to get an HDR film of the billboards.
For the driving green-screen sequences of the film, we had built a plate to mount three cameras, at 0, 45, and 90 degrees, to shoot panoramas driving down the street. To get the nodal points closer together, we had the cameras facing toward each other, with the lenses almost touching. It worked wonderfully.
By taking the center camera out, and replacing it with a beam-splitter, we had a down-and-dirty HDR rig using the other two cameras. Now, this was HDR on film, not video -- but film already has a very high dynamic range -- so two cameras with very different effective exposures gave us a tremendous dynamic range. In the 'normal' exposure all of the brighter signs were blown out, but on the beam-splitter camera you could see all the details of the structure of the lighted billboards. Quite cool.
I love Mondays. On a Monday, anything is possible.