A7s: S-log2 too noisy? Then underexpose cine4

Perhaps I wasn't clear enough with my English - I meant that when you need to properly expose you main subject, dark against the bright sky (which Cine4 is intended to handle best), it's much easier to preserve the sky details (blue sky, white clouds) with S-log2 than with Cine4. Which is great - but with the same S-log2 that makes it possible to preserve the blue sky - banding in the sky is inevitable, which sort of denies the advantage of preserving it :(

Piotr
 
I'm not getting any banding in those conditions. Are you using a LUT? I did get ugly banding and unnecessary noise when I tried a LUT-based workflow (then concluded there was something wrong with my workflow, and went back to my current curves-based workflow).

I shot most of this video with subjects in shadow and blue sky in the background, using slog2 on an RX100 IV. No lights, not even a reflector. I think the highlights look great, and I see no banding on those skies.
(warning: kpop music!!)
https://www.youtube.com/watch?v=eGp_JIAUE_s
 
No Samuel - I don't use LUTs; I much prefer Resolve Color Management, sometimes ACEScc. I hoped the A7Rii would make a good, competent B-cam for my FS7 - unfortunately, its 8bit footage is very limiting - especially given that I grade for HDR10. When I found that out - I bought the GH5 which is now a true B-cam for my video work; the A7Rii is an excellent stills camera for me... But I'd like to squeeze the best possible quality out of it - sometimes I only have it on me, and something I just couldn't afford not to film happens :) Thanks,

Piotr
 
Last edited:
Well, no idea then. I don't have banding issues, but I don't use resolve, and I use some very particular settings in the camera (decent level of saturation from the start, see here), and also maybe my bar is lower (i.e. you'd see banding where I don't).
 
moldcad said:
Perhaps I wasn't clear enough with my English - I meant that when you need to properly expose you main subject, dark against the bright sky (which Cine4 is intended to handle best), it's much easier to preserve the sky details (blue sky, white clouds) with S-log2 than with Cine4. Which is great - but with the same S-log2 that makes it possible to preserve the blue sky - banding in the sky is inevitable, which sort of denies the advantage of preserving it :(

Piotr
Click to expand...

In that case you may have more luck using an external recorder to get 8 bit 422 out of the camera instead of internal 8 bit 420.
But the DR is so impressive with my cine 4 settings I rarely find skin under exposed unless severely back lit.
 
I suppose you could experiment with dithering to alleviate the banding. In this video, Matt Scott concludes that banding won't appear under normal grading circumstances, as long as you don't use a LUT. It's curious that you would have the opposite experience. I don't think that Matt showed any examples of exported clips with heavy grades applied, so that may be why he experienced different results than you.
 
moldcad said:
I've read this thread with interest, as an A7Rii owner for just couple of weeks. From my own experience so far, nothing I do to Cine curves (especially Cine4) can beat what S-log2 is the best at: ability to preserve highlights (like the blue sky with white clouds), even if I overexpose the S-log2 by 2 stops to bring it don in Resolve and get rid of some mid-and-low area noise.
Click to expand...
Quite simply, if you can't preserve the highlights, you are overexposing. Decrease the exposure and you increase your highlight range. Of course then it helps to have a workflow that supports exposure compensation, e.g., shooting raw, using ACES, or using my Logarist workflow, which supports Cine1 and Cine2. I think you'll find that S-Log2 has no advantage over Cine1 or 2 when used with a color managed workflow supporting exposure compensation.


HOWEVER - and I wonder why nobody even mentions it in this thread - the 8bit nature of A7 series cameras make S-log so prone to banding (like in the blue sky), that I rarely use S-log2 anymore! Can you explain to me why blue sky (or white wall, for that matter) is not banding at all (provide of course that it's not totally blown out) when using Cine gammas, and banding like hell with S-log2, practically making it unusable ? TIA

Piotr
Click to expand...

In S-Log2, from one 8-bit code value to the next is a much bigger color jump than in Cine1 or Cine2. S-Log2 was designed for 10-bit cameras, and doesn't perform so well in 8 bits. Cine1 and 2, on the other hand, perform very well in 8 bits, with a distribution of code values that's much closer to the final output. That means less banding. S-Log2 and 3 are useful on cameras with better dynamic range and higher precision recording - not the Alpha-series bodies.
 
bill totolo said:
In that case you may have more luck using an external recorder to get 8 bit 422 out of the camera instead of internal 8 bit 420.
But the DR is so impressive with my cine 4 settings I rarely find skin under exposed unless severely back lit.
Click to expand...

Bill - I AM recording to Samurai Inferno; 4:2:2 color is not enough though for the 8-bit S-log2 to become banding-free...

Piotr
 
Imamacuser said:
I suppose you could experiment with dithering to alleviate the banding. In this video, Matt Scott concludes that banding won't appear under normal grading circumstances, as long as you don't use a LUT. It's curious that you would have the opposite experience. I don't think that Matt showed any examples of exported clips with heavy grades applied, so that may be why he experienced different results than you.
Click to expand...

Yes, I often use dithering for alleviating the banding problem, but it can only help so much :(

Piotr
 
moldcad said:
Thanks @balazer - will study your Logarist workflow with interest :)

Piotr
Click to expand...

OK, so I tried your workflow on my slightly under-exposed Cine1 footage; delivering to 709 gets quite nice results. However, I tend to grade most of my FS7 S-log3/S-Gamut3.Cine footage for HDR (using either RCM or ACEScc). Do you think if I kept your Cine1->Logarist LUT, but replaced the Logarist->709 one with a proper 2020 output, I'd get a better HDR picture form my Cine1 A7Rii footage that could get close to the FS7 output? If so, did you create your Logarist output LUTs in the 2020/2084 version? TIA

Piotr
 
Hi, Piotr. At the moment I don't have any Logarist output transform for ST 2084 with BT.2020 gamut. I will try to build one and get back to you. I can also make an ACEScc transform. In theory, yes, with the right color space transformation it could be close to FS7 output.
 
Thanks Jacob! Please let me know when you have created such output LUTs, or ACEcc transforms (with the latter, I guess input ACEScc transforms would be necessary as well, because we would then work with Resolve's ACEScc and not YRGB color science), am I right? TIA

Piotr
 
Being relatively new to Mirrorless A7III settings, having worked most of my life working with cameras based on a SMPTE based Matrix I though I would hunt around for a heads up on the A7III. Rather than digging around into the camera and spending hours running my own tests I though I would try and get a leap in on where to start with any settings adjustments. I discovered this video on the Tube and it gave me better than a head start. I though I would try his suggestions and after a couple of days trials I basically have to concur with his findings. 90% of my A7III work is based on HLG3 2020 if I'm grading the material myself and HLG3 with 709, also requiring some curves grading but the color is not as flat as the 2020, if I'm handing off the images. So far I'm finding these by far the quietest gammas plus they allow you to use lower ISO settings than SLOG2's 800. Finding these two settings with minor changes here and there suitable for most jobs and so far so good.

https://www.youtube.com/watch?v=D2iqEDUe1qg

Another one of his videos regarding the internal 420 vs 422 external recording I found to be pretty accurate re the A7III.

https://www.youtube.com/watch?v=K4586b8iR2Y&t=488s

Chris Young
 
Last edited:
You must log in or register to reply here.
Back
Top