Can a colorist explain highlight recovery to me?

[sander kamp]

First I wonder if this a good way to check if there are recoverable highlights in an image:

In RedCine uncheck the 'color' tab at the top, or in RedAlert set output to LUT to 'linear light'.

In that way I am looking at the raw, uncolored footage - the linear output from the sensor, right? So any bright area that is not blown out still contains information that you can do something with in post?

My confusion is now if this last bit of information is really useful or that in order to make nice roll-offs of the highlights you need more room, let's say another stop? It seems to me that in order to use the last bit of highlight you would need to treat highlights in the image separately from the mid-tones and shadows, is it normal to do two passes out of a raw file to do so?

Two images out of RedCine, first with REDspace color profile applied and blown-out sky, the second one with 'color' button unchecked and lots of detail in the sky. Now how to get the mid-tones right again? Whatever curve I apply in RedCine I can only get either one of them right.

 

[albert rudnicki]

this is calling for a secondary and masks.
Go color

 

[Kyle Presley]

try an S curve, maybe? I'm no expert, but I can usually get good results from video footage that way.

 

[Daniel Reichenbach]

Or develope it twice, one for the blacks, one for the whites, then mask the versions together.

 

[Nook Kim]

try an S curve, maybe? I'm no expert, but I can usually get good results from video footage that way.

With that much of contrast in the scene already, I don't think applying an S-curve will help in anyway.

Sanders,

I'm not sure what gamma setting you're using, but it looks like Redspace to me. I'd suggest you to try using Redlog or PDLog 685 and back down the exposure setting a bit (usually around -2). Then you could apply an S-curve to adjust the overall contrast to your liking. If you have a CC tool with proper video out, I'd recommend applying an S-curve there. I can't still read the Histogram in RA as well as I do waveform. But this is me.

See if this improves your result. If not, Daniel's way sounds reasonable as well.

 

[sander kamp]

Sure, I understand the options of masks and developing twice. But I would like a experienced colorist to explain if this is a viable option or that there might be problems when doing this. I work regularly on the set as a DIT and occasionally find myself working with DP's who like to expose with highlights very close to overexposed. It is good for the shadows and to keep noise under control but I wonder if the highlights might be a problem in post or that there is an easy way to handle this in a Scratch, Da Vinci, Nucoda etc. system.

 

[sander kamp]

With that much of contrast in the scene already, I don't think applying an S-curve will help in anyway.

Sanders,

I'm not sure what gamma setting you're using, but it looks like Redspace to me. I'd suggest you to try using Redlog or PDLog 685 and back down the exposure setting a bit (usually around -2). Then you could apply an S-curve to adjust the overall contrast to your liking. If you have a CC tool with proper video out, I'd recommend applying an S-curve there. I can't still read the Histogram in RA as well as I do waveform. But this is me.

See if this improves your result. If not, Daniel's way sounds reasonable as well.

Hi Nook, thanks for your answer. Here I get a little confused. Is applying a different color space not the same as applying a different curve setting? Or is it something completely different?

 

[Bruce Allen]

My 2c: don't try to do it in RedCine - it just doesn't have the power (power windows that is . And fighting with the strange settings there will drive you nuts.

Just render it out twice - once with settings to make the sky look good, and once where the cityscape looks good.

Blend these in the program of your choice. If I was using After Effects, I'd put the one clip on top of the other with masks. If I were in Avid I'd put one on V2 and the other on V1, and use an AniMatte (where you can paint the transparency of a layer with soft paint tools). If I was using Final Cut Pro I'd cry. Don't even bother with their mask tools, they suck. Maybe you could get away with a simple grad in this case, or painting something in Photoshop? Take it into Color instead if you're going the Final Cut Studio route.

Bruce Allen
www.boacinema.com

 

[John Tissavary]

Here I get a little confused. Is applying a different color space not the same as applying a different curve setting? Or is it something completely different?

A curve, as used in color tools, is nothing more than a paradigm for a user interface allowing control over an (theoretically) infinite number of points of the luminance of whatever component of the image is applicable (r, g, b or any sum thereof in the case of RGB images). Since curves are mathematical representations of one-dimensional spaces, they often represent one-third of a typical three-dimensional color space.

Color space is an abstract color model mapped to a specific values. These values can be words, numbers, or a combination thereof.

In our biz, the purpose of choosing a color space is to pick colors that will fit inside the abilities of the target display mechanism (film, CRT, LCD, etc...). Colorspaces you may have encountered include Rec709, sRGB, CMYK, HSV, P3, ad nauseum...

So what does all this have to do with which you should use for your image? F-all, I say. If it looks good, use it.

As Bruce mentioned - if you're using RedCine, then rendering both versions and combining them with masks in the compositing tool of your choice (AE, Shake, Nuke) is a very effective way of getting all the dynamic range you seek out of a single image. If you'd rather do it in a color tool like Color, Scratch, Baselight, etc... then you can do similar operations there, except what you're really doing is using a single image and creating masks to direct certain color operations to specific areas of the image - it'll save on render time if nothing else.


cheers,

JT

 

[Adrian Sims]

I'm no colorist but I'm getting great results using Color Finesse from within my NLE, which allows you to isolate your highlights and drop their brightness with little effect on the midtones and lower ranges. I tweak the following settings after I crush my blacks with a Color Finesse S-Curve and apply broadcast safe limiting (soft clip enabled) and that usually does the trick. The color adjustment filters that come with most NLEs are OK, but in my opinion don't cut the mustard when compared to 32-bit color filters like Finesse. Works for me.

 

[Eki Halkka]

That doesn't look too hard a shot to do even in Redcine. I took the darker jpg and applied a simple curve in photoshop, it worked decently with 10 seconds of tweaking ;-)

(See attachment)

 

[Álex Montoya]

Yeah, it's not that hard. It may be easier for some of you to first apply levels to get the midtones higher and then apply curves. I would try something like this since it looks more natural, but if you want a darker sky just apply a mask... all the detail is still there:

If you're working at 16bits you shouldn't have problems with banding, etc...

 

[Nook Kim]

Is applying a different color space not the same as applying a different curve setting? Or is it something completely different?

Hi Sander,

Color space is a certain range of color gamut, which is different from gamma setting. Theoretically, gamma setting is like applying different curves.

Just to clear your thoughts a bit more, Redspace has harder gamma value (more contrast if you will) than Rec709, and Rec709 has harder gamma value than Log. That's why Log gamma looks way more washed out compared to Redspace. What I was suggesting earlier was to choose the easiest gamma value, and put your own curve instead of just using Redspace and limit yourself from the get go.

Hope this helps.

PS. I came across this whitepaper by Autodesk that discusses this subject a little bit in another thread:
http://www.postmagazine.com/Media/MarketingManager/Autodesk Red whitepaper final.pdf

 

[Jeff Brue]

also the purpose of a linear light pipeline is also for CG in which case linear compositing methods through a look gamma is always the way to go. In regards to pipelines the red doesn't have enough sensor depth to justify going with a two exposure strategy as a 16 bit linear light tiff will contain all of the information that would be in those two exposures. Technically a LOG exposure should as well but, that could be subjective.

From a workflow perspective doubling your data to achieve the greatest image latitude isn't realistic, For problem shots ust go to 16 bit linear light tif ad apply your gamma and secondary color corrections until you get what you want.

 

[Jeff Brue]

oh ant btw only the 10 bit log dpx route will have any hope of being realtime in todays hardware.

 

[Paul Leeming]

First I wonder if this a good way to check if there are recoverable highlights in an image...

Can you post the R3D for us to work with? I think a reverse S-curve would be a good start, anyway.

 

[Kevin Shaw]

Colorist opinion

Colorist opinion

Hi Sander
Well I am a colorist, so here are my thoughts.
First, the best way to check if there is highlight detail is to look at Redspace: Redlog. The color space settings are more than just gamma curves in the sense that they should provide a known start point. I prefer redlog because it can show all the dynamic range of the capture on a monitor at the same time.

Your issue here is essentially one of dynamic range. The camera captures more dynamic range than a monitor of "normal" gamma can display. Thats why any of the linear settings require correction.

Once we establish the camera has captured the highlights, the question is how do we grade the image. With my colorist instructor hat on I would describe 3 stages.
1 -Primary. Because film also captures more dynamic range than video all pro color systems have some sort of separate controls for black, mid and white detail. Some are better than others, and some separate more than others, and some have different calibrations for linear and log sources. For example a da Vinci 2k just has linear black, mid and white controls. A Nucoda film master has 3 sets of black,mid and white, each of which can be calibrated for video, CGR or log. One of them (brightness regions) also allows you to define the white threshold and would therefore bring back the sky with minimal effect on the street

2-Isolation. Use a combination of luminance key and graduated shapes to darken the sky. This is more powerfull, but moves away from what was shot. So the question becomes is it a correction (for the dynamic range) or an enhancement (to add more impact from the sky than was shot)

3. Blend modes and LUTs. These are new tools to us colorists, but they are a simple way of using primary tools (1 above) and isolations in a way that retains more of the DPs intent than the isolation techniques above.

Hope that helps
Kevin

Sure, I understand the options of masks and developing twice. But I would like a experienced colorist to explain if this is a viable option or that there might be problems when doing this....

 

[Shawn Booth]

Thank you Kevin.