FAQ S-log2 vs. S-log3 for HDR grading: which is better?

moldcad

moldcad

Veteran
As we know, S-log2 can utilize about 98% of the 10-bits of data available while S-log3 is only using about 87.5% of those 10-bits; the latter assigns more info below the grey mid-point of 18% and allows for circa 1.5 wider DR. But grading for HDR in Resolve, I've been wondering whether using S-log2 would actually be more beneficial by retaining more bits of information in the highlights - even if with lesser overall latitude. What do you guys think?

Piotr
 
Last edited:
broughtonfilm said:
Looks like another 100 posts and no decision thread to me...
Click to expand...

Instead of your little patronizing brickbats, please try to to provide some theoretical knowledge to help me (and - I'm positive - many others, using FS7 footage and trying to achieve the best possible results in HDR grading).

Ben Scott said:
I think go and shoot something and test for yourself :)
Click to expand...

Oh, but I did - I experimented with the same scene shot and recorded with both S-log curves, while trying to keep all other things equal as much as possible. But since it was a "quick and dirty", far from scientific, experiment - I was unable to come to unambiguous conclusions. Hence my need to get some theoretical background from those more experienced than myself before I adopt the apparently obvious and simple "receipt" : use S-log2 in very bright scenery, and S-log3 in more subdued lighting scenarios...

Because for the highlights to really "stand out" in the final HDR grade, it's not enough for them to be recorded at higher IRE (yet without clipping) by the camera (which definitely is easier using S-log2). In order to separate them and thus make more dramatic, the midtones and shadows must be graded darker yet still retain as much information as possible (lots of shadow details and color information, and all that without excessive noise). And this in turn can be achieve much easier with S-log3, so we're having two contradictory goals here...

A simple example: with S-log2, I can easily make the bright sky with super-white clouds reaching well over 1000 nits in my Resolve scopes, but cannon darken or flatten shadows without losing lots of subtle gradations and details in this area, additionally degrade by the excessive noise (getting rid of which using NR removing detail even further). While on the other hand, with S-log3 recording of the scene, I'm able to keep shadows and midtones flat (in the 0-100 nits range) yet detailed and virtually noise-free - but am unable to pull highlights much above 1,000 nits without losing those white/grey clouds again blue skies; they all turn into a very bright mush without detail...

I realize that even with the same overall dynamic range of 12-14 stops that the sensor is capable of, if I recorded 12 or 14 bit RAW I could probably retain detail in both extremes, but hey - I'm discussing Sony's S-log here, as implemented on the FS7 semi-professional camera...

Piotr
 
Allister Chapman & I had an interesting conversation about this back at NAB. We had concluded (and his & Sony's later testing appears to prove out) that the best way to shoot HDR on a camera such as the F55 is to shoot SLog3 overexposed by two stops. The easiest way to do this is with a two-stop overexposure compensation LUT applied and then expose "normally" based on this LUT. The idea is to push those highlights way up there into the top of the range since that's what HDR opens up so much for you. Then the rest trails off into the bottom of the range the most efficiently. We got into the conversation because we had been discussing how effective and useful the preset Exposure Compensation LUTs are that we created for use in the Odyssey7Q+. And it turns out that they work perfectly for HDR as well.
 
I have to admit that I don't know what you mean by "HDR grading" to begin with. People refer to many things when they say HDR. HDR understood simply as an image of a wide dynamic range of 12-14 stops but made to be viewed on a regular Rec709 display, HDR as the same image but to be viewed on new generation displays capable of displaying wider dynamic range, or HDR understood as a stills photo technique where multiple exposures are stacked (to create kitschy images)?
 
Thanks Mitch - but it's exactly ETTRing S-log3 which leads to clipped-out superbrights. It's much safer in S-log2, but in this case shadow detail is suffering.

So basically, S-log3 overexposed by 2 stops is a good method for reducing shadow noise, but only for Rec.709 final product, IMMHO...

Piotr
 
Last edited:
broughtonfilm said:
I have to admit that I don't know what you mean by "HDR grading" to begin with. People refer to many things when they say HDR. HDR understood simply as an image of a wide dynamic range of 12-14 stops but made to be viewed on a regular Rec709 display, HDR as the same image but to be viewed on new generation displays capable of displaying wider dynamic range, or HDR understood as a stills photo technique where multiple exposures are stacked (to create kitschy images)?
Click to expand...

I mean HDR as defined by the "HDR10" specification (the workflow basically identical to the "proper" Dolby Vision one, but limited to some 1,000-1,500 max brightness as allowed with "Premium UHD TV" specs).

Piotr
 
moldcad said:
Thanks Mitch - but it's exactly ETTRing S-log3 which leads to clipped-out superbrights. It's much safer in S-log2, but in this case shadow detail is suffering.

So basically, S-log3 overexposed by 2 stops is a good method for reducing shadow noise, but only for Rec.709 final product, IMMHO...

Piotr
Click to expand...

It's not overexposing and clipping stuff. It's exposing correctly and putting the image where there is the most information available to it to manipuate afterwards. Seems a bit weird too ask for advice and then get an expert tell you best practice and then you tell him he's wrong.
 
I used the IMHO keyword. Meaning I'm not saying the expert is wrong - merely that the advise doesn't seem to work for me.

Piotr
 
Ben Scott said:
But you shouldn't be clipping anything.
Click to expand...

OK - you're right, Ben; when I try to boost the highlights into the 1,000 nits region in Resolve, my problem is not clipping as such, but not enough information in the highest brightness portion of the picture.

Donny 123 said:
If I were you..I would pack in camera work altogether and become a writer.. :)
Click to expand...

I already am a writer - no need for packing in my camera work. But thanks for your advise, my good man.

Piotr
 
Oh c'mon Ben - what you call "Mitch's advise" has been touted by Alister Chapman ever since S-log became popular, and I did follow it precisely for the reason Alister was stressing - getting rid of shadow noise by bringing everything down in grading. But I don't want to bring everything down when grading for HDR - can't you see this subtle difference?

Piotr
 
Why do you need to bring everything down?

The idea is to move the image into the area where there is the most data allocated for it during acquisition and to space it out correctly in post. So bring the shadows and mid tones down to where they should sit and space the highlights where they should go. Just about any correction tool can do this quickly and easily. If you decide you have a set formula you like you could even build a preset HDR transform lut to do it for you

This is completely different to simply exposing as far to the right as possible.
 
Ben, I'm grading precisely the way you described (apart from using LUTs - RCM is better IMHO :). And as I stated, with S-Log3 footage I'm lacking information in highlights, while with S-log2 - in shadows... Well, I guess this is so much that can be done with 10 bit footage, even if theoretically 14 stops wide in DR... Thanks,

Piotr
 
In my experience with the FS7 Slog3 looks much better and grades easy. I usually push exposure a bit and for most of the time I end up with a great looking image with enough room in post to fine tune highlights and shadows.

S-log3 isn't the answer to every kind of shoot but for most of the stuff I shoot it works great. I did a comparison Slog2/3 almost two years ago and concluded Slog3 worked best and never looked at Slog2 for the FS7 since.
 
moldcad said:
Ben, I'm grading precisely the way you described (apart from using LUTs - RCM is better IMHO :). And as I stated, with S-Log3 footage I'm lacking information in highlights, while with S-log2 - in shadows... Well, I guess this is so much that can be done with 10 bit footage, even if theoretically 14 stops wide in DR... Thanks,

Piotr
Click to expand...

If that is the case then surely it is either down to how you are exposing or how you are colour correcting.
 
You must log in or register to reply here.
Back
Top