Understanding HDR gamma

Erik Wittbusch Nov 15, 2016

  1. Hi,

    setting up for my first small HDR test project.
    I want to check, what the PQ EOTF of HDR does and how the project in Resolve need to be set.


    I set up a UHD project for HDR10-1000nits first.

    Here's my settings in a DaVinci YRGB Color Managed workflow:

    HDR_settings.jpg



    When I put a traditional 10 steps on the timeline, the scopes look like this:

    steps_ST2084.jpg

    I just thought, that a HDR10-1000nits EOTF will limit my max. brightness automatically to 1000 nits and be logarithmic.
    Obviously, this ain't true. Only the output is set to ST.2084 - not the footage.

    What do I need to do for a technical correct transfer now?



    I need to put a technical SDR(gamma of choice) > ST2084 1000 nits 3DLUT on the 10 steps.
    This is what it looks like:

    steps_realST2084.jpg

    Not bad, but quite far away from hitting 1000 nits!


    Let's adjust the max brightness to 1000 nits then:

    steps_realST2084gain.jpg

    Ahhh!
    That's at least how I imagined it to look.
    But is it also supposed to be like this?

    Any help highly appreciated!
     

    Attached Files:

  2. I think you have been tricked by the ramp you generated. If I recall correctly it changes behaviour with every different output color space chosen. Try to not use it, but use a simple image that you might have already graded to REC709 SDR. Or first create a proper SDR ramp.
     
  3. It might be that RCM doesn't tone map between colour spaces, try using ACEScc and see if anything changes.
     
  4. It's the 10 Step Generator from Resolve and it should be okay.

    What is a 10 Step ramp supposed to look like if it's correctly transferred to ST.2084?
     
  5. Okay.

    Next try.
    I made a grey-ramp in Resolve (REC709, gamma 2.4) and exported as UHD 16bit-TIFF.

    Then:
    1) imported into the above mentioned project (RCM: REC709 input + timeline color space, ST.2084-1000nits output color space)
    2) put the REC.709-2.4gamma to ST.2084-1000nits LUT on top
    3) and adjusted gain for max. brightness to reach 1000 nits

    Here's the Waveform image:

    grey-ramp_RCM_REC.709-ST.2084-LUT.jpg
     
  6. I set up a ACEScc project with the following settings:

    ACES_REC.709-ST.2084_settings.jpg


    Then imported the above mentioned REC.709-2.4gamma grey-ramp and put in on a timeline.

    Here's the waveform:

    grey-ramp_ACES_709-IDT_to_ST.2084-ODT.jpg

    I thought it should look identically even without adjusting anything as I'll set up Resolve for the same transform. Or am I misunderstanding something here?
     
  7. I think you are applying the transform twice. By selecting st2048-1000nits in output color space you are already applying a transform so there is no need to put the lut on top.
     
    Erik Wittbusch likes this.
  8. Resolve's Rec.709 input transform is an inverse of its Rec.709 output transform, including the RRT tone mapping. It is designed to transform an incoming image so that when viewed through the Rec.709 ODT it will look the same (pretty much – the RRT is not 100% invertible) as the original. The inverse tone map is what creates the curve you see there on the waveform. It is exactly what I would expect.
     
    Erik Wittbusch likes this.
  9. So it does look right.

    If I use a REC.709 graded clip instead and put it in the same timeline, the output should be a technically correct ST.2084 HDR footage, right?
     
  10. Depends what you mean by "correct". The only purely technical transform from SDR to HDR is one which produces an image on an HDR monitor which peaks at 100nt, and therefore looks exactly as it would on an SDR monitor. Creating an image which has highlights above 100nt requires stretching out the compressed highlights of the SDR image, which is what you see at the top of the curve in your ACES waveform. There is no one "right" way to do that. The ACES approach will give you one possible start point, and you grade from there.

    But obviously, the best way to make an HDR image is not to start with an SDR one!
     
  11. Thanks!

    But what can you do if your client wants to see how the SDR footage looks in HDR formats before deciding anything further?

    Of course, starting with 16bit RAW or some LOG-encoded footage would be better.
     
  12. Sure. If your SDR material was graded in ACES, by working in ACES with Rec.709 input transform and ST.2084 output transform, you'll get close to what the original log footage would have looked like with the ST.2084 ODT. If it wasn't graded in ACES, you will be inverting an output transform which is not actually what was done to the footage. Nonetheless it will hopefully give you a reasonable start point.
     
  13. But what If it wasn't?

    I did receive some footage where nobody knows where it really comes from.
    I assume that it's standard REC.709 - gamma 2.4 graded footage.

    To check how the same footage will look in HDR, it really should be regraded in HDR with the original footage. Agreed 100%.

    But to give the client an idea, I thought I can just adjust the levels, sothat the max highlights hit 1000nits
    in my ACEScc project with REC.709 IDT and ST.2084 ODT.

    Actualy I don't have a HDR capable display and therefore can't really grade the footage.
    I can only adjust the max brightness while trying to keep the average brightness on the waveform with no visual control.

    This is ridiculous, but I'm still waiting for my display. It will hopefully be here within 2 weeks.
     
  14. Creating HDR without an HDR display to view on is, as you say, ridiculous! If I really had to do it, I would just stick to the straight Rec.709 IDT + RRT + ST.2084 ODT. You still wouldn't know what that looked like until you saw it on an HDR display. But making "blind" adjustments, simply to max out the 1000nt would be a really bad idea.

    An SDR image may well have flat clipped highlights, and just pushing those to be flat areas of 1000nt would not look good, and might well cause automatic power saving gain reduction on the HDR display.
     
    Marc Wielage likes this.
  15. Thank you Nick, that helped a lot!

    You should really give courses for deeper understanding of color management

    Unfortunately, that's how I thought, too.
    I was hoping for some easy transfer without artistic choice that makes sense nevertheless.
    But how can you grade blind...
     
    Paul Dore likes this.
  16. I am quite a rookie at grading HDR, though I lost my virginity yesterday, after the arrival of my new client display, which is hdr capable. Something I realized after playing around a bit is that most of the HDR goodness is actually found in the SDR highlight regions. For example, normally the highlights on somebody's face and specular highlight coming from a lightbulb behind him/her all lay very close to each other in a SDR graded image. But in HDR this is were the fun starts and where you can really create more depth by separating those specular highlights a bit further from the facial highlights. So to start with an SDR graded image where those highlight regions are often compressed isn't the best idea. Also just brightening the image till 1000 nits also doesn't make much of a difference. It's playing around in that extra range that creates a cool effect IMO.
     
    Adam Hawkey and Erik Wittbusch like this.
  17. I just found some SLog3 footage with direct sunlight to play with.
    Now, that makes a hole lot of sense!

    There's no need to compress those spectacular highlights - in a heartbeat I'd say that's how it's supposed to look anyway.
    The thing is, that we all learned to compress to BT.1886 color space and really get used to it...
     
  18. Hi,

    just another thing:

    When I export graded footage from the ACEScc project with the ST.2084 ODT, it should already be with the HDR/PQ EOTF and mapped into REC.2020 gamut.

    I did so, exported to ProRes and then compressed to H.265 10bit.

    The people from LG told me, the TV didn't recognize HDR or REC.2020 signals, but mapped it to SDR REC.709.

    What went wrong?
     
  19. The file needs to have the appropiate metadata/flags to trigger the tv into HDR mode. I think a 2.0 HDMI connection is also mandatory for this to happen.

    How did you create the H.265 file?
     
  20. Doesn't Resolve export the flag?

    I imported the ProRes into a special 10bit Handbrake version and encoded a UHD H.265 in 10bit.

    Then I used the Youtube HDR-flag tool to make sure it's flagged.

    Didn't work...

    Do you have a different workflow?
     

Share This Page