P3 to 709 workflows


(Charles Boileau) #1

Hi to all,

With this post I’m hoping that someone will be able to give me some simple insight into using ACES to produce different deliverables.

I’ll first establish a context: I’m grading a trailer in DCI-P3 in Resolve 12.5.5.

At the end of my session I decide to produce the 709 version by switching the ODT to 709. Online needs to have a 709 version quickly. Which is not our usual workflow. Normally, I would render to P3 and convert to 709 with a LUT after online. At the moment, it’s not possible to online and package in ACES within Resolve. Titles look awful, transitions and fades don’t behave properly etc…

I change rooms and look at my new version in 709. Looks pretty good, but I’m in a hurry. I don’t think much of it. A few weeks later I see the trailer on my computer and notice it looks a little bit washed out and dull. I get the same comment for the DP…

Yesterday I have a few minutes to look into this. So I go ahead a compare different P3 to 709 conversions. Honestly, I thought the going from P3 to 709 would be more seamless. But it’s not… It’s “very different”.

I’ve already had this discussion already by email with the ACES folks. From what I understood, a colorspace conversion LUT is a “straight up thing”. The ACES “conversion” is setup to have a intent. But what is the intent? What am I missing here that I don’t understand? Why are the images dull and flat? And why is the white point bluer?

The Rec709 (d60 dim.) looks a little bit better. It seems to have the correct WP. But still is flat and dull…

So please can someone explain simply (not science lingo) what we’re supposed to do with this? Are we supposed to adjust the grade? What is the intent that ACES but into this conversion? Is the fact that we calibrate our monitors to 2.2 the issue. Or, am I just not understanding what this is and what I’m supposed to do?

Cause from what I read ACES was supposed to be real simple. Need a 709 output? Just switch to 709…

Here are some examples:
Split screen…
On the left side: Grab from my ACES project. IDT applied is Rec709.
On the right side: Grab from a P3 render (from ACES) with a P3 to 709 LUT applied.

On the left side: Grab from my ACES project. IDT applied is Rec709 (D60 sim.)
On the right side: Grab from a P3 render (from ACES) with a P3 to 709 LUT applied.
As you can probably see. The WP is much closer to my original.
And I’ll also state that the P3-709 LUT is spot on.

Thanks for the input!


(Scott Dyer) #2

The “Rec. 709” ODT that ships with v1.0.x will not match the P3 version in the way that you are comparing them.

The Rec. 709 transform is designed for a display calibrated to 100nits, D65 white, following the Rec. ITU-R BT.1886 EOTF model (gamma 2.4) for viewing in a “dim surround” where an observer is adapted to D65.

Let’s ignore the white point difference for now and just talk about what else is different.

To compensate for the dim surround, the transform has a gamma reduction factor because in general, less contrast is needed when rendering for a brighter assumed viewing environment.
Also, because Rec.709 monitor is assumed to be brighter than P3 cinema (100nits vs 48nits), the transform also includes a desaturation matrix to compensate for the Hunt effect (which basically says that colorfulness increases with luminance).

And obviously if your monitor setup deviates from the calibration and environment assumptions stated for the Rec. 709 transform, then you won’t get the intended picture.

The P3-to-709 LUT would be appropriate for when your Rec. 709 monitor is set-up in the same environment as where you did your P3 mastering - i.e. dark surround. (That is, assuming that the LUT is doing what I think it should be doing)

If ACES v1.0.x supplied a Rec. 709, D60sim, dark surround ODT, I would expect that to pretty much achieve what you’re after with the P3-to-709 approach.

EDIT: Apologies if too technical. Perhaps some others can weigh in and comment from a different perspective.


(Jim Houston) #3

Scott nailed it. I would add that the P3 and the 709 outputs were visually compared
so that the images had an apparent match when comparing a projected image at 48 nits
and a Rec709 image at 100 nits with a D65 dim surround around the monitor.

I would only add that, in my opinion, there is an adjustment in the adopted D65 ODT that shifts all of the
neutrals from D60 to D65 because of viewer adaptation, but since many viewers watch these shows in the dark, I prefer to use the D60sim version which is for calibrated D65 monitors so that the output master does not have the white point shift. If they are watching in a daylight condition, it will seem a bit warm, but cooling the image sometimes changes what a colorist chooses to do with the white point, and warm rarely hurts.

The emphasis was on the deliverable end point, not so much what would happen in direct comparison
on an editorial monitor. There has been feedback about this which is being discussed for ACES 2.0 –
it comes down to, what is a ‘match’ and when and where do you look at it. Any thoughts?


(Nick Shaw) #4

Not quite. Because P3-DCI and Rec.709 have different white points, even a “straight” conversion transform has a few options. Do you apply a chromatic adaptation matrix or not, and if so which one (Bradford and CAT02 being probably the two most common).

The idea of a chromatic adaptation is that it converts the image so that the colours in the Rec.709 version should appear the same to a viewer whose eye/brain combination is adapted so that they perceive equal Rec.709 RGB values as being white, as they appeared to a viewer who was adapted to perceive equal P3-DCI RGB values as white when they watched the P3 version.

But whether you want chromatic adaptation applied or not is somewhat dependent on whether when you graded the P3 you did so working to equal RGB on the waveform being white, and let your eyes adapt to the slightly greenish DCI white, or whether you offset the white in the grade as a creative choice to take it somewhere closer to D60 or D65.

But that does not account for the effect of brightness and viewing environment on perceived contrast and saturation, as Scott described above.

Conversion is complex, and there is no one “right answer”. The idea is that you want to watch the Rec.709 version on a BT.1886 monitor and have it “feel like” the same film you saw projected. That does not mean that if you put a video monitor in the screening room it should look identical to the colours on the projection screen.


(Charles Boileau) #5

Thanks guys! You all bring very valid points. And these are all things that I’ve considered when trying to understand how ACES converts from one colorspace to another. And it’s all very relevant and logical.

But that being said… What is a colorist supposed to do when he sees the teaser on youtube and says: “God, it looks like crap… It’s light years away from what we did on the projector.”? This is a reaction I never had when using straight conversion LUTs.

I want to use ACES. I think it’s the future… So please give me practical info. And I know that sometimes the ACES folks have to be nuanced as you are not aware of what our setup is.

  • Should I adjust the grade based on a memory I have of the project in P3? Cause it really did look like crap…
  • Should I revise how we calibrate our grading monitors (who by the way are all calibrated to 1886 gamma). But we’re 125 since we feel that most TVs and displays are usually brighter that 100nit. It’s a compromise that we’ve induced on purpose.
  • About our gamma… We use the 1886 gamma. But in Lightspace it says that the curve (when we analyse it) is at 2.2. What we’ve always been using. So, again please indulge me… COuld we stick with 2.2 or should we move to 2.4? 2.2 has always treated us well.
  • When going to the web… Should we use the Srgb ODT? I remember that it seemed like a bad choice (from what I logically knew at the time about conversions).

I’m really looking for day to day insight. I know that ACES is working hard to make things technically correct with a whole bunch of variables…

Thanks!


(doug_walker) #6

Hi Charles -

I agree with the excellent posts from Scott, Jim, and Nick. (And they also give you a sense of the technical discussions we have on the ACES committee.)

However keep in mind that the gamma adjustment factor Scott mentioned (to compensate for the surround difference) is less than 2%. In your case this is overwhelmed by the fact that the Rec.709 Output Transforms (ODTs) you’re using are intended for 2.4 gamma and it sounds like you’re looking at it on devices that are either calibrated to 2.2 or sRGB.

BT.1886 is essentially a 2.4 gamma, sitting on top of the device black. If you are mastering for traditional Rec.709 video delivery, then this is what you should be using. The standard for web delivery is sRGB (which is also called “RGB Monitor” in the ACES transform collection). You also mentioned YouTube and that one is harder to specify since people watch it both in web browsers and on smart TVs. It would be interesting to hear from the community what gamma people feel works best when mastering for YouTube (I’m guessing web is still the best choice).

To sum up, the practical advice I would give is as follows:

  1. Calibrate your devices according to the delivery format you are mastering for. (ACES cannot help you with this part.) Use BT.1886 (essentially gamma 2.4) for TV and sRGB for web. Note that sRGB is very different from gamma 2.2 in the shadows and the ACES transforms really use the sRGB curve, not 2.2.

  2. Choose the Rec.709 Output Transforms when mastering for TV and the “RGB Monitor” (or sRGB) Output Transform when mastering for web.

  3. Choose the “D60 sim” flavor of Output Transform if you want to preserve the native cinema D60 whitepoint in your video deliverable, or choose the other flavor (without “D60 sim” in the name) to adapt D60 to D65 so that it will sit more comfortably with standard TV or computer content (e.g. commercials, computer user interfaces, etc.).

Hope that helps,

Doug Walker
Autodesk


(Charles Boileau) #7

Thanks @doug_walker! Great post! I’ll look into it further!

Thanks to everyone for the help!

Cheers


(Charles Boileau) #8

Hi to all,

I’ve been looking into this particular transform more closely. And I’m now seeing the utility of having the 709 odt behave the way it does. The D60 sim also seems to bring me to a happier place.

That being said… I wanted to push my tests further and look at the sRGB D60 sim for web deliveries. My initial post was motivated by a DP seeing his teaser on Facebook which he thought looked pretty flat and dull.

I exported a new version of the teaser with the sRGB D60 sim ODT and low and behold, it looked pretty good all over the place (iPhones, browsers etc…)!

This move really rocked our “internal workflows world” as we never thought that we needed to go to sRGB (709 being so close).

On another recent project I had a very dark and very saturated scene (heavy blue cast). It looked pretty awful once compressed to H264… Especially with the 709 transform that boosted the gamma “upwards”. It looked like a pile of block pixels that we’re all the wrong colors. Tried the sRGB D60 transform and it pretty much cleaned up the problem. But this was on my grading computer… There was obscure forces at work!

I decided to have a look on other computers, my iphone etc… This particular scene seemed to work only on my grading UI monitor. iPhones, iMacs and others looked crushed and plain disgusting.

So does anyone have any insight on this? I know the web is like a jungle. You never know what you’ll get… But i’d still like to give a clear explanation to my clients. Is it graphics card that don’t get updated? There basically stuck on the old 1.8 sRGB gamma and curve?

What are others doing about this?

Thank you very much!

Cheers


(Charles Boileau) #9

No one has an opinion or insight on this?


(Nurali Kushkov) #10

Hey there,
I’m researching on the same subject and find answers here insightful.
I just have one comment:
I quickly brought in a test image from arris website and flip through P3 DCI, Rec 709 and Rec 709 60sim. White point on DCI P3 and rec 709 60sim is not similar at all, to my eye at least.

Oh and one more thing, which bugs me and I’m very curious to hear from experienced folks:
I’m watching scopes while flipping through those output transforms, scopes change too.
Why is that?
Shouldnt neutral remain neutral on scopes???
thanks in advance!

DCI P3, resolve 14.0

Rec 709 60sim

Rec 709


(Alex Forsythe) #11

@k.nurali If I understand what you’re doing, this is the expected behavior. Both the P3 DCI and the Rec.709 D60 sim ODTs have a “D60sim” in them, however, because the calibration white point of the devices is different the code values required to produce D60 on the screen are different. That’s being reflected in the waveforms. The Rec709 ODT does not have a D60sim so objects in the image with equal ACES values will have equal levels on the waveform. It will reproduce on the screen as D65 because that’s the calibration white point of the Rec.709 device.


(Nurali Kushkov) #12

Hi Alex,
thanks for your reply!
Ok, i see what you mean.
I have one more question then.
I just finished colouring a short film, which i did on Christie 2k projector in DCI P3.
I now need 2 masters, one for DCP and another sRGB for web.
Thing is finalising will be done later, so i wanna avoid double work for final editor.
He’ll assemble the film together, opening titles, sound, end credits, graphics etc.
Could i render dpx stack in p3, have all finalising done and bring that master file back into resolve for p3->srgb 2.2 pass?
Or would you suggest simply render both (p3 and srgb) and do 2 separate master files?
Thanks for any ideas.
Best regards,
Nurali


(Charles Boileau) #13

To the best of my knowledge you should render both from your project. Bringing back the P3 and converting to SRGB will most likely give you an unwanted result (wp, gamma etc…). You should try it to see the difference.

Have you had good results exporting to SRGB? I can’t seem to wrap my head around the conundrum of SRGB. Most projects will look much better (overall) if I use the 709 ODT for web.

This was the original point of this post and still no one has given me some insight on this.

Thanks!


(Nurali Kushkov) #14

Hi Charles,
thanks for your reply.
I decided to stick to Aces workflow and render both deliverables instead of messing with it. Finishing editor will simply relink to another DPX stack i render for web.
I never used SRGB, actually, only REC 709. It works ok, i dont think its spot on, but close enough in most cases, as viewing environment is so different from grading room.
I didnt have time for extensive tests yet, only did a quick one for a project i delivered recently and found srgb profile being actually pretty close to what i see in my grading room, when i compare open project on my ref monitor with srgb render on my laptop. But more on a dark side when laptop is out in the “wild”, day lit office room.
Planning to do more tests when christmas fever is over, probably in december.
Cheers, hope you’ll find your answers soon enough :slight_smile:
Best,
Nurali