What are the 'correct' skintone values in IRE for HLG acquisition?

[Mmmbeats]

I want to know what the typical reference values are for skintone exposure in HLG for standard dynamic range delivery.

I'm finding this information quite easy to source for linear and log curves, but a bit more murky when it comes to HLG.

I am aware that -

• Skintone exposure is subjective, and can vary according to artistic intent, mood, lighting, contrast, etc.
• Skintone references are a bit out of date because camera sensors have a wider DR than when they were formulated
• The whole thing is a bit euro-centric and often doesn't consider the diversity of skin tones available

However, I like having a set of values (ideally skin tone and skin highlight tone for lightest skin types, and the same for darkest skin types - I then guestimate for tones inbetween).  It helps me to sanity check, and acts as a foundation for my subsequent choices.

In other words, I'm not intending to slavishly expose every thing to the same skintone value - I just like to know where I stand!

I'll be shooting HLG on GH5 and GH5S.

For linear, I usually work to these values (for lit skin itself) :

Lightest skin: 60 - 65 IRE

Darkest skin: 45 - 50 IRE

For V-Log-L I have found these values (which are for skin highlights) :

Lightest skin: 55 IRE

Darkest skin: 42 IRE

For HLG the only info I've found is the following:

Lightest skin at 1 stop over - 78 IRE 

This is as per Gerald Undone. I like the guy a lot, but I don't quite trust his workings on this one.  He seemed to be using 18% grey at 50 IRE as his middle reference, but I'm pretty sure it's supposed to reference at 38 IRE for HLG.

Asian skintone (slightly darker than lightest skin) - 50 – 55 IRE

As per TL;DR Filmmaker.  He's another youtuber I've got a lot of time for, but I'd classify him as an 'intelligent amateur' rather than a pro.  It's pretty clear that his understanding of HLG generally was all over the place at the time of doing the tests.  Having said that, the resulting skintones looked good.

Lightest skin: 60 IRE

This is per Alister Chapman, who I would definitely trust above the previous sources.  Only thing is, it was just a fleeting mention in a video interview, so I'd like to see it confirmed, and have a way to elaborate other skintones and values.

One thing I do notice is that the upper end of darkest skin values seem to coincide with the 18% grey reference point:

 Linear: 50 IRE

V-Log-L: 42 IRE

So for HLG this would be: 38 IRE (as per ITU-R report)

Perhaps that could help in determining part of the range?

Anyways, apologies in advance if I've mangled any of the concepts or values above.  I'd like to know typical reference values for HLG, basically,

[Mmmbeats]

Righto - so I've spent this evening doing some tests using my Colorchecker Video chart to try to hunt down some answers to this.

Just in case anyone else is seeking this info, here are my findings.

I monitored Natural, V-Log L, and HGL.  Each white balanced custom, at f/2.8, 400 ISO, using shutter speed to vary exposure.

I used reflective white as a reference point for each profile, and then used the middle grey chip as a second reference.  I had to make a best guess as to which of the very similar light grey chips was 18% grey as it is not noted in any of the XRite documentation weirdly enough.

So -

The reference profiles gave readings similar to what was expected for lightest skin:

Natural: 70 IRE

V-Log L: 50 IRE

The reading for HLG was:

HLG: 57 IRE

So, reassuringly close to Alister Chapman (60 IRE), and TL;DR Filmmaker (once you adjust for skin tone).

The readings for darkest skin were much lower than expected:

Natural: 19 IRE

V-log L: 29 IRE

HLG: 18 IRE

This leads me to think that either:

1) The darkest skin tone chip on this chart is particularly dark

2) People have not calculated exposure to include darkest skin tones properly

3) I've made some kind of mistake

Overall, I'm pretty satisfied with the test results though.  The lightest figures correspond well with how I understand each curve.  I'm willing to believe that darker skin can fall along that part of the exposure range.

I hope these are useful to someone else.  I can post more about my methods (which admittedly weren't perfect), if anyone needs the info. 

 

 

[hyalinejim]

9 hours ago, Mmmbeats said:

I had to make a best guess as to which of the very similar light grey chips was 18% grey

I don't know for sure, but if X Rite says that the second large chip is "40 IRE" then surely it must be 18% grey.

[Mmmbeats]

51 minutes ago, hyalinejim said:

I don't know for sure, but if X Rite says that the second large chip is "40 IRE" then surely it must be 18% grey.

That's the one I used. Its labelled as 40 - 50 IRE on the documentation I have. 

[hyalinejim]

So if you expose that chip so it sits at 42 IRE on the in-camera waveform for VLog-L then that should be the technically correct exposure.

Then you can note where the relevant skin tone chips fall on the waveform, and switch over to HLG to cross reference the values.

That said, I've found that using an 18% grey reflective target as a guide to exposure gives varying results, for a few reasons.

Firstly, the angle you hold the chart at in relation to the light source will raise or lower the brightness of the grey patch quite a bit. Say you've got one key light at 45 degrees above and to the right of the subject. You'll get significantly different readings depending on whether the chart is facing the camera or angled towards the light source.

Secondly, in bright contrasty sunlight a grey card exposed at 18% gives results that are far too dark, to my eyes. 

The more you look into things like this the further you go down the rabbit hole! So I think that grey card metering will get you within a stop or two of where you need to be, but I would always adjust to taste. 

How about spot metering on skin using the camera's exposure meter? It should work the same regardless of picture profile. Of course, in contrasty lighting skin won't be uniform either... I don't think there's any perfect method of exposure. Even incident metering gives varying results if there's backlighting. 

I would love to be able to perfectly expose though. It would save that little bit of time in post.

[Mmmbeats]

20 minutes ago, hyalinejim said:

So if you expose that chip so it sits at 42 IRE on the in-camera waveform for VLog-L then that should be the technically correct exposure.

Then you can note where the relevant skin tone chips fall on the waveform, and switch over to HLG to cross reference the values...

Yes, that's more or less the technique I used, though middle grey has a slightly different reference point for HLG (38 IRE is what I used).

The way I usually do exposure is by looking at the talent and figuring out where I want their skintone relative to 'normal'.  This is influenced by whether or not they are in shadow, whether it's a moody scene, the shade of their skin, etc.  I rarely drop it more than a stop below 'normal'.

Once that is established, I take a look at the highlights and whether they need protecting.  If they do, I make some kind of compromise - either I allow some elements to blow out, or I allow the skin tone to drop further than I would like.  This just depends on the aesthetic of the shot (for example I might let quite a big window blow out, but I would hardly ever let a big section of sky blow out).

Finally, if I think there is important stuff going on in the shadows I might make an adjustment to try to help them a bit (while making sure to continue to protect the highlights).  I very often skip this stage and just let the shadow area fall where it may.

Sounds long winded written down, but all of that happens very quickly and automatically.  I prefer to use false colour, but am happy to use zebras if that's all I have access to.  As you say, spot metering is a really good alternative too (though you have to work out over/under values to make it work on most of the cameras I use).

No way am I faffing about with a grey card - I want a technique that works the same across events, doc, fiction, etc. (obviously there are somewhat different approaches required for each).

If the skintones are in the ballpark of where you want them in the final grade then I find the post route is a lot more straightforward.  You're also less likely to damage them by pushing or pulling.  That's another reason why I prefer them as a starting point for exposure than middle grey.

Having said that, I do sometimes push my exposure up a bit if I have the headroom - a kind of ETTR-lite! (the coward's version 😉).

I don't really think that 'exposing properly' means always providing end goal exposure levels in-camera.  It's virtually impossible to preserve both highlight detail and ideal skin tone exposure in a large percentage of shots, unless you were to shoot on an Alexa or something (out of my budget range!). 

 

[hyalinejim]

I also do a lot of run and gun type stuff and don't have time to be faffing about with getting perfect exposure. I take a glance at the exposure meter (with the camera pointed towards the ground rather than the sky), the zebras (set to 80 so I can see what's actually clipping in the file) and sometimes I switch on the waveform for a quick look at that. And I also base a decision on the LCD or viewfinder (it's quite easy to see if things are overexposed but underexposed can still look fine, so it can be misleading).

So I end up with a variety of exposure levels but I'm usually within one stop. Sometimes it's two stops, which is pushing it a bit, but that's life!

When it comes to colour correction, having something two stops off in V-Log can be a real pain in something like Lumetri in Premiere. The exposure slider does not behave like, say, the exposure slider in Lightroom when adjusting RAW. So the best way around that I've found is using an ACES workflow in After Effects, where as if by magic you can reliable make RAW-like (ie; rational and consistent) adjustments to exposure, contrast and white balance. And if I'm feeding the signal into a good LUT, then that's all I need.

It's an extra step in the workflow though, so usually I don't bother. Or don't need to, if I've already got exposure and WB quite close while shooting, so that Lumetri doesn't fudge things too much. I imagine that in Resolve it's possible to make proper adjustments to Log or HLG (so that if I underexpose 2 stops I just need to add +2 in post and I get the same image, but noisier, than if I had exposed correctly)... but I haven't made that move to Resolve yet.

But it's a real pleasure to be able to do that when needed. Here's an example of the same scene shot at correct exposure with daylight WB, and then 3 stops under with coolest WB. This would be a nightmare to correct with conventional methods but with ACES it's very, very close. There is a red tint in the darkest areas of the corrected underexposed shot, but I think that's due to red chroma noise being boosted.

[Mmmbeats]

4 hours ago, hyalinejim said:

I also do a lot of run and gun type stuff and don't have time to be faffing about with getting perfect exposure...

Those post routes look very worthwhile.  I still haven't taken a look at ACES, and don't even have a working understanding of it yet.

Is there a benefit to reading the ground for exposure?  I've not heard of that one before.

I'm increasing seeing (from sources I trust) that 75 IRE zebra is the best clipping warning level for V-Log L because of the risk of individual channel clipping (at 80 IRE).

[hyalinejim]

2 hours ago, Mmmbeats said:

I still haven't taken a look at ACES,

One of the huge advantages of working within ACES is that gaps between stops are equally spaced. Each stop is the same distance apart on the waveform. So it's truly logarithmic (conventional log curves have a significant toe in the shadows) - each doubling of light results in a boost of the same number of IRE units. This makes it incredibly easy to colour correct as any effect that globally affects the waveform will correct your footage as if you were making adjustments to linear RAW data.

Here's twelve stops of VLog-L straight out of the camera:

 

And here's that same step wedge in ACES CC colour space:

Brightness, contrast and WB adjustments now become easy and accurate. If you're trying to pull something out of the toe of a conventional log curve it's a disaster as the contrast is different to the linear part of the curve. So in that situation where you're metering roughly and your WB is a bit off this workflow will give matched shots that are identical except for noise in underexposed shots and highlight clipping in overexposed shots. BUT it's a pain because it's an extra step. If Premiere could do ACES I would do it there. And for sure you can already do it in Resolve.

Incidentally, it's also now very easy to see when things get noisy. How many clean stops for the GH5 shown here? I count 9!

2 hours ago, Mmmbeats said:

Is there a benefit to reading the ground for exposure? 

This is just based on me noticing that when I'm on evaluative or center weighted metering the thing that makes the meter go wild is having sky in the shot. 18% grey is supposed to be the average reflectance of the average scene, so I think that the ground is much closer on average to 18% than the sky is (although clear north sky is supposed to be 18%.... not where I live lol!) So things like concrete and grass are quite close to 18%. So what I'm talking about here is pretty much just looking at the meter as I take the camera out and switch it on to give me a rough ball park of where I should be, then fine tuning it with zebras, waveform, LCD chimping etc. This is all very much in the same run and gun spirit of what you're talking about - just trying to get shots that are within a stop or two of where they need to be.