HDR in REDCODE RAW

J

jeffry alan

Member
There is a feature of CS2 where a raw image is processed for gamma, one for highlights and one for shadows, then those 3 exposures are combined into one image to reveal a high dynamic range (HDR).

How feasible would that concept be for REDCODE RAW?

J*
 
Well, you've got raw data, and can certainly do that if you want. However, I'm not precisely certain what difference you'd achieve via that method than a well constructed curve. You can't do as definate a blend as you can with a still as that would certainly produce artifacts on motion.

Graeme
 
I think with the curves editor you are still limited to the prevalent dynamic range, and HDR broadens the range to a point where the details in the shadows and highlights can be brought out, (a curves editor for each of the highlight, midtones, and shadows, and then blended) creating a greatly extended latitude.

This would seem to be a possibility with the increased definition offered at 4k, and in the workflow would precede color timing and would essentially be like a motion control setup which shouldn't introduce any artifacts.

J*
 
Perhaps - but I'm concerned about artifacts from this on motion. Also, such HDR blending does produce issues with halos which don't look quite right - how will they appear on motion? I guess this is a little R&D project for you :-)

Graeme
 
Depending on which algorithm one uses to create their HDR, halos can be defeated entirely.

I think it would be an interesting test... why not?

What's the word on taking a single CR2 file and turning it into an HDR? Is there some automation script in CS2 that'll do it for me? I've used the assemble to HDR automation before... and I can imagine doing it from 3 developed JPG files from a single CR2... but is there a way to automate that process?
 
Not familiar with the details of the HDR function in CS2, I have been working in CS and read about HDR in Stephen Johnson's latest book, although it is common enough to layer (at least) three RAW exposures optimized for highlights, mids and shadows, and blend in Photoshop, trying to match Ansel Adams zone system approach to printing as I do (which creates a ginormous file size), before creating a final PS, TIFF or jpg.

I think you can use the 'automate/batch' feature and choose the commands for every image, however I think you would want to have manual control over all of your exposures.

I like to do all of my panoramics myself (matching each of the red, blue and green channels of each photo seperately) rather than rely on the PS panoramic stitcher to automate the process, which tends to glitch up on my pano's which are usually longer than ten images stitched together.

J*

p.s. Brook, did you work on Pursuit of Happyness?
 
Last edited:
traditionally HDR captures are a function and generated off of f-stop or shutter speed though limiting and defining the amount of light viewable to the sensor. I think at the end of the day all you're seeing in that wikipedia image is a tone mapping operation applied to a 12 bit image.

It can create some astounding imagery though, but it's nothing you could use in a more standard use of HDR like generating 3d lighting information... however I'm looking forward to the day when to capture on set light... all we have to do is hold up a chrome sphere in front of the camera.
 
If you are generating a single HDR image from a single RAW image, regardless how many separate steps you take, there is always going to be a single (possibly complicated) function that does the same thing. If each pixel is independent of the others (eg. no blur or USM type operations), then that function can be expressed as a 1:1 mapping, eg. a well constructed curve, as Graeme mentioned.
 
and the fact that things are actually moving makes HDR an issue, as graeme states you must introduce interpolation between movement and that causes some effects.
 
Jeff Brue said:
traditionally HDR captures are a function and generated off of f-stop or shutter speed though limiting and defining the amount of light viewable to the sensor. I think at the end of the day all you're seeing in that wikipedia image is a tone mapping operation applied to a 12 bit image.
Click to expand...
But instead of locking down the camera and shooting multiple takes with different exposures, (which is essentially the same but labor intensive), you take a single RAW file and create three or more versions with exposures ranging from shadow detail to highlight detail and combine them.

J*
 
Jarred Land said:
and the fact that things are actually moving makes HDR an issue, as graeme states you must introduce interpolation between movement and that causes some effects.
Click to expand...
It's nothing that hasn't already been adressed in traditional motion capture systems for the last thirty years.

Are you saying it can't be done, or it hasn't been done yet?

J*
 
it can be done.. but to know how HDR works.. well traditionally , is that it takes x amount of frames of the same still object and combines them. Which was great for still buildings and landscapes , but as soon as you add movement it all goes to hell.. unless you shoot all 3 images within a nanosecond of each-other... which with high speed cameras is possible (at around 5000-frames per second) but you have to remember you also need to adjust the shutter in that time as well. Motion control as suggested would work on still things, but not for moving people.. or moving anything, unless that anything was also motion controlled.

Bottom line.. is that its a SERIOUS pain in the ass to do HDR in a video camera.. and there is no cost effective benefit since most of the time it can be done practically with light control. If you push HDR too far.. it begins to look non-natural.
 
Why not just work in floating point?

HDR is great for CGI work because it brings that much more realism to a scene, especially since most of the lighting is designed.

I would guess an camera designed for HDR would have to be designed like a 3 strip Technicolor camera, one CMOS each for gamma, highlights and shadows.

This would mean 3x the data for each shot.

Someone may be doing it, but like Jarred said, how cost effective will it be.

If you want to know what the master of HDR is doing, go to www.debevec.org/

I saw his latest lighting rig that they are using for research with video, but they are using a high speed camera and have lights set up in a semi-hemisphere geodome configuration.
 
jeffry alan said:
But instead of locking down the camera and shooting multiple takes with different exposures, (which is essentially the same but labor intensive), you take a single RAW file and create three or more versions with exposures ranging from shadow detail to highlight detail and combine them.

J*
Click to expand...

That sounds suspiciously like cutting the file into three parts and then glueing the tonal ranges back together... ie. useless.

You aren't going to get *new* data out of this process that wasn't already in the original file. If you apply gain to the upper end, to simulate a higher exposure, youc ould have done that on theo riginal using a curve. If you apply gain to the shadows to simulate a longer exposure, you could have done the same thing using the curve... my point is... all you're doing is waving your hands and throwing smoke to end up with the same product you started with.

It sounds like a lot of people here are expecting to get 8bit RGB images out of their raw files. If REDCine artifiicially mapped the dynamic range to a 16bit file (also useless) you would have a half float HDR image. It would be a simple matter of adjusting a histogram and saving as a 16 bit image to achieve what HDRshop does. If you then mapped that luminance to a radiance map, you now have a 'true' mathematical HDR image. No combining/splitting needed, just a declaration of "this is 100 lum" "this is 5 lum"
 
The method I am suggesting is a modern day "Technicolor", in that there are three IDENTICAL prints. One negative (RAW), three+ print exposures combined.

J*
 
Shaw said:
Why not just apply a curve to map all that tonality into your output?
Click to expand...

Because you are still limited to a normal dynamic range (6-9 stops). The definition of (H)igh (D)ynamic (R)ange is that it expands that range significantly (12+ stops), and a simple curves editor would have to be expanded as well to reflect all of the increased tonal range.

Any adjustments on a curve editor would be gross, unless you could do macro adjustments to each tonal range or 'zone'.

J*
 
Last edited:
Splitting and recombing cannot work. Whatever dynamic range you captured is the all you have to work with. Setting up the exposure and controlling the lighting correctly will do far more to make a shot work in challenging lighting situation. The dynamic range produced by combining multiple exposures is orders of magnitude larger dynamic range than any camera I have seen. Well there was a test rig that I saw at ICT in or near Paul Debevec's lab that used 6 nd filters and a prism to simultaneously capture multiple exposures, but I never saw what came out of it. There is a lot of research in this area. Just following a few links on debevec.org leads to several other universities working on HDR video.

To add to the pot of HDR capturing for the Red:
My vague idea for how to get something HDR-like would be to sample the data off the chip more frequently than the actual exposure. For instance at 1/60th, you could also pull out the data at 1/120th and 1/240th and 1/480th and so on. I don't think it would work too well, but it is fun to think about this kind of stuff.

Of course the real solution is to build chips with much higher dynamic range (Mysterium 2?). For now working with limits is part of the challenge. And for those who do not like the limits, the challenge is getting rid of those limits.
 
Jarred Land said:
If you push HDR too far.. it begins to look non-natural.
Click to expand...

I'm sorry to stay this, but I can't see why RED (or any digital cinema company with a healthy sense of self preservation) would ever want to supply shooters with this capability - not until people start demonstrating some restraint with HDR & Tone Mapping techniques. Yes, used subtly HDR can be a powerful tool, and it has some wonderful applications in post (like environment maps for 3D or "light probes" for image-based lighting) .... but for the love of Christ most of these putrid HDR landscape "auto comps" doing the rounds make me want to hurl. Why would RED want to be associated in any way with such monstrous, nightmarish imagery such as this:

http://en.wikipedia.org/wiki/Image:Ladder_and_Cherry_tree_08077.jpg

Cool - no more shadows - kill me now...

Notwithstanding the technical hurdle of "baking" multiple exposures within the same time slice without butchering the shutter speed, I humbly think that the better way for digital cinema manufacturers to proceed is to keep building better, more sensitive and less noisy sensors with greater native dynamic range and higher bit depth or floating point. There is also some promising research being done by an Australian company which offers an advanced nd filter which effects only the highlights of a given scene (sorry, lost the link). Until that time I will be happy to wait for the inevitable HDR backlash, when people will be less inclined to indulge in shameless gimmickry, and will hopefully be more concerned with creating artful and natural imagery. End rant. :)
 
Last edited:
You must log in or register to reply here.
Back
Top