FS7: The Great S-Log-2 vs S-Log 3 Debate
- Thread starter karma17
- Start date
morgan_moore
Major Contributor
Which is why I was saying that Mr Jensens claims were off (in terms of evaluating FS7/5-XAVC) as he is taking from the raw source from his F55
Doug Jensen
Carbonite Member
I have a R5 and two 512MB cards that I'd like to sell if anyone is interested. I upgraded to an R7 awhile ago and have not gotten around to selling the R5. It's in perfect condition and no blemishes.
Doug Jensen
Carbonite Member
I stand by what I said. I am not wrong and I don't think you understand how RAW works in Resolve.
BTW, I believe that what Tom is saying is that RAW enables one to mix and match gammas and gamuts in any way a user wants to, so S-LOG2/S-Gamut-3.Cine is possible.
morgan_moore
Major Contributor
Lets look at 10 stops below clip on Den chart.
Slog3 recorded at 10bit gives you 38 levels of grey
Slog2 recorded at 10bit gives you 17 levels of grey
The 38 levels can take a load more grading before breaking than the 17 levels.
In fact 10 stops below clip Slog 2 has basically turned to mush.
But record in 12 or 16 bit and you have a square power more data (or some other big mathematical jump)
Lets guess Slog2 at 'raw 10 stops below clip' records 70 (not 17) steps of data while Slog3 100 steps (both highly gradable)
Now sling your raw into Resolve and apply whatever transform you want and you are not having to waddle around in 17 level mush becuse you recorded either 4000 tonal values (12 bit) or 65000 tonal values (16 bit)
Resolve is not destructive in raw - I know that much - after having deeply questioned the production team on exactly that point.
Last edited:
Doug Jensen
Carbonite Member
Believe what you will, I do not have any interest in debating it.
You said it better.
16 bit raw also has more shades at -10 stops than either SLog2/3. X-OCN is really the way to go, if you shot enough with X-OCN what you would save in archival storage would pay for the more expensive recorder and media. There isn't that option with the FS7 but an F5/R5 combo can be found at a very affordable price.
Last edited:
Doug Jensen
Carbonite Member
I love XOCN. All the power of RAW, but at manageable data rates. I only have a single 1TB S48 card and it serves me very well for 4K @ 120 fps. The $4500 price tag was hard to swallow, but it's more than paid for itself on stock footage alone already. Everything from here on out is gravy.
morgan_moore
Major Contributor
We are all in agreement:
raw/ocn is way bettter than Xavc.
raw/ocn can be shot in Slog2 Slog3 and wongled in post to look the same
Where Doug and I, I think, disagree is that lessons he learns about profiles when shooting raw are transferable to lower bit depth cameras.
Pretty important when one is an 'instructor' for lower bit depth cameras.
----
Ive been there; In 2008 when transitioning from stills (16bit raw digital hasselblad) to my first video camera (EX1) I applied all that I had learned about exposing a 16bit raw hasselblad file to exposing my EX1.. it messed me up for years
raw/ocn is way bettter than Xavc.
raw/ocn can be shot in Slog2 Slog3 and wongled in post to look the same
Where Doug and I, I think, disagree is that lessons he learns about profiles when shooting raw are transferable to lower bit depth cameras.
Pretty important when one is an 'instructor' for lower bit depth cameras.
----
Ive been there; In 2008 when transitioning from stills (16bit raw digital hasselblad) to my first video camera (EX1) I applied all that I had learned about exposing a 16bit raw hasselblad file to exposing my EX1.. it messed me up for years
Last edited:
Doug Jensen
Carbonite Member
Not only important, but also true. That is why I said whatever it is you are referring to this time.
morgan_moore
Major Contributor
What codec?
So you have performed the first step
And that step involved you making a choice. 'I could see the image.. but it was too noisy so I didnt count it'
To noisy for what? To look like a 'glossy advert', a 'gritty documentary' or 'stand up in court as evidence' image?
Id suggest if your 'bar' is 'glossy advert' you may have culled a couple more stops*. (or my theory.. 2 from Slog3 and 3 from slog 2)
The second step would be to shoot something tonal (maybe a 1/3 lit grey sphere) at each of those IRE values and try grading the image of the sphere. Ie pushing it around a bit in post, adding contrast etc.
At that point you might decide that 'that stop' is not suitably robust in the grade.. and cull it.. probably losing a couple more stops (with XAVC) leading to my DR figures of 7 and 9 for doc and 6 and 5 for glossy.
Of course doing that test so formally would be time consuming, a more practical approach could be to shoot a big sidlit sphere and get a feel for that thrown around in a grade.
When I test I shoot an old beauty dish cross lit to get a nice tonal surface.**
*if you shot the XAVC codec which is what I understand this thread to be about the 'cull rate' would be lower for raw I guess
**my F5 and FS7 tests are not available online any more as I dropped Dropbox. Ill have to re host them.
So you have performed the first step
And that step involved you making a choice. 'I could see the image.. but it was too noisy so I didnt count it'
To noisy for what? To look like a 'glossy advert', a 'gritty documentary' or 'stand up in court as evidence' image?
Id suggest if your 'bar' is 'glossy advert' you may have culled a couple more stops*. (or my theory.. 2 from Slog3 and 3 from slog 2)
The second step would be to shoot something tonal (maybe a 1/3 lit grey sphere) at each of those IRE values and try grading the image of the sphere. Ie pushing it around a bit in post, adding contrast etc.
At that point you might decide that 'that stop' is not suitably robust in the grade.. and cull it.. probably losing a couple more stops (with XAVC) leading to my DR figures of 7 and 9 for doc and 6 and 5 for glossy.
Of course doing that test so formally would be time consuming, a more practical approach could be to shoot a big sidlit sphere and get a feel for that thrown around in a grade.
When I test I shoot an old beauty dish cross lit to get a nice tonal surface.**
*if you shot the XAVC codec which is what I understand this thread to be about the 'cull rate' would be lower for raw I guess
**my F5 and FS7 tests are not available online any more as I dropped Dropbox. Ill have to re host them.
Last edited:
abreu-canedo
Well-known member
The gamma curve is a non-linear way of encoding and decoding luminance values.
I'm not technical enough to understand the fine details of how bits are distributed between RAW capture and it's application across different codecs.
In XAVC world, the bit distribution is part of the gamma conversation. But is the gamma affecting bit distribution in RAW as well? After all, you can decide how to interpret the RAW signal through your choice of gamma curve in post, no matter what gamma you choose to monitor in on set.
I also agree with the comment above for the subjectivity of dynamic range, as well as the various factors that can determine the interpretation thereof.
I'm not technical enough to understand the fine details of how bits are distributed between RAW capture and it's application across different codecs.
In XAVC world, the bit distribution is part of the gamma conversation. But is the gamma affecting bit distribution in RAW as well? After all, you can decide how to interpret the RAW signal through your choice of gamma curve in post, no matter what gamma you choose to monitor in on set.
I also agree with the comment above for the subjectivity of dynamic range, as well as the various factors that can determine the interpretation thereof.
None!
So you have performed the first step
And that step involved you making a choice. 'I could see the image.. but it was too noisy so I didnt count it'
To noisy for what? To look like a 'glossy advert', a 'gritty documentary' or 'stand up in court as evidence' image?
I asked the most high for strength and victory, she said to count the angels dancing on the head of a pin.
Dynamic range is a state of mind.
How are we supposed to calibrate our own eye's?
https://www.cambridgeincolour.com/tutorials/cameras-vs-human-eye.htm
How are we supposed to calibrate our own eye's?
https://www.cambridgeincolour.com/tutorials/cameras-vs-human-eye.htm
You have the full understanding already about the distribution of bits, the only point is sensor voltage is linear light, you need gamma transformation (SLog/2/3) so that you can store enough shades when there aren't enough bits. With 16 bits, you have enough shades without a gamma transform, so any gamma you would choose within your NLE is an arbitrary choice.
Using 10 bit XAVC may cause you to consider one instead of another.
I've been following along this wonderful old debate... and enjoying most of the banter. And to Tom's / Sam's point about 10bit XAVC and the scene's dynamic range and bit value per stop might making you choose one instead of the other in camera.. I also wonder shouldn't final "delivery" also be a consideration?? If we're squeezing our wonderful camera's 14 stops (ok, let's realistically say 12) down to a final output of say 6 or 7 stops for non HDR delivery...isn't the tonal quality of each stop and subtle gradations effected more by that compression?
The tonal quality and subtle gradations are affected by compression, but not DR. It's considered 6 or 7 stops for non HDR delivery for one or both of the following two reasons:
1.) DR is scene light. To show 14 stops of scene DR you need a display device with 14 stops of actual brightness.
2.) Recorded Camera 709 is so highly compressed above the knee, it's likely not possible to see shades above 6 or 7 stops.
But you can *represent* the shades above 6 or 7 stops, maybe all of them, (even if wrongly counted!) with gamma transfer. All you have to do is show recorded SLog/2/3 on a 709 monitor. All 14 stops are represented within the image. They are compressed however by the log gamma to fit within the brightness range of the 709 display device so they will appear flat, but the details are present. If you de-log it though, into display gamma so that it shows with a normal amount of contrast, than details will have to be highly compressed where you may not see them or clipped unless it is HDR.
morgan_moore
Major Contributor
Id say the opposite is almost true.
Consider a screen trying to show a 6 stop scene over 10 bits (1024 levels - call it 1000 for simplicity) that is 1000/6 levels per stop. 166 levels.
So if you had a gamma curve set on your camera that filled the 'container' (0-1000) when subjected to a six stop scene you would collect 166 levels per stop.
Now if you recorded the same six stop scene with Slog you would be collecting somewhere between 48 and 113 levels per stop or worse if you under exposed.
You are seriously short of data here!
So the risks are high when using Slog that you will get a worse image than when using a baked codec.
Really the only reason that (10bit) Slog works at all is because mainly we deliver 8 bit 256 levels. Slog 10bit is really a non starter for any delivery medium that is more demanding.