Are S-LOGS More Destructive Than They're Worth?

[maxotics]

42 minutes ago, Deadcode said:

If we push these images waaay further and we increase the saturation and the contrast the SLOG2toREC709 footage will have more banding and more color noise compared to REC709. But not that much more than you think. 

In conclusion: yes SLOG2 gives less color information and you should not film in that profile every time (if we talking about 8bit)

But in real world filming the benefits of using SLOG2 is worth over it's weak spots.

 

I don't know how many times I need to say that I too believe all LOG recording gammas have their place in the filmmaker's toolkit.  A really experience/serious filmmaker is probably not going to use a camera limited by 8bit video, as @cantsin has pointed out and can use a LOG gamma in a scene near rec709 and not worry about it too much.  For one to say, "in the real world filming the benefits of using X___ is worth over its weak spots" is a subjective judgement, right?  The benefits may be worth it to you, not to someone else.  We can never know what we'll want when we shoot.  I'm trying to figure out the differences between LOG and non-log shooting in the A6300 for example.  

I've done tests similar to the one you suggest, but i wouldn't be telling anyone here anything they don't already know!  There are some disputes about where the noise comes from, but no one disputes that LOG shooting introduced noise somewhere.  And most don't feel the noise is a big problem.  

Aren't you even a little bit interested in the JUST HOW MUCH mid-tone information is lost in LOG.  Even if it's small, even if no matter what the number, you're going to keep shooting it, wouldn't you like to know?  I would.  Unfortunately, it's proving very difficult to get at that data.  I've spent hours writing scripts and developing techniques but am hitting technical difficulties which many try to help me solve by saying, 'don't bother, we know everything.'  I'd be fine with that if they shared their statistics.  Just tell me the numbers.  How much noise is introduced by various LOG recording gammas?  How much color information is lost in the mid-tones in LOG.  How aggressive will a LOG need to be to introduce visible banding?  

 

[Bioskop.Inc]

These LOG problems/misunderstandings seem to be appearing a lot recently. I've read through most of the thread & there has been some really useful information put forward, some useful links (more or less). The main point that keeps coming to mind is, Has anyone actually tested the DR of their camera with & without the LOG profile? I know this might seem such a basic thing, & might be insulting to some, but it is the thing that is screaming out to me if you want to know how a LOG profile bahaves.

You read time & time again how professional DPs test cameras, so perhaps this is what you need to do to understand how the LOG profile behaves & so understand how to use the LOG profile. Simply put, the easiest way to do this is to test the DR of the camera with & without the LOG profile.

The following quote came from another forum & was suggested by David Mullen (a professional DP) as a simple way to do this:

"Put a white card and a black card on each side of an 18% gray card -- under and overexpose in whatever increments you want until you can't see a difference between the white and gray card at the overexposed end and between the gray and black card at the underexposed end. Make sure you shoot at whatever is the widest dynamic range recording format the camera offers (raw, log, cine gamma, hypergamma, etc.) but also test it in the narrower Rec.709 display gamma range just for comparison."

You can, of course, do this anyway you want & test all sorts of different aspects of the camera - you can test how different ISO/ASA behave, how different F stops behave etc... I'd suggest that you use a Histogram & Peeking (make notes of the values at which the camera over exposes/under exposes - all of them). You can use ND filters, but make sure if you use a Vari-ND that it has markings.

You can, and considering the topic at hand, should add a colour chart to the test - but if you do this White Balance the camera to the white piece of card (with BM cameras you'll need to test the various WB presets). You should also try to make sure the Black card is as black as possible - "The black card should be as black as possible, some people create a recessed box lined in black velvet".

Just remember that not all recording formats are equal e.g. 8bit vs.10bit or H264 vs. ProRes (do the test with what you'll be using). And do the test using your equipment & how you intend to shoot.

Having done this simple DR test you should now know how the LOG profile of your specific camera behaves, compared to the preset profiles (Rec.709). If you make careful notes then you'll know what to do with your camera in a variety of situations.

Finally, please remember LOG isn't RAW & you don't need/have to use either of these recording profiles to get great images!

Oh, and sometimes you might have to make a choice & crush blacks or blow highlights.....

[cpc]

Testing camera profiles comes with the job characteristic of being a videographer/cinematographer. The easiest (cheapest) way to get an idea of DR distribution is to shoot a chip chart at a few exposures with a constant increment and use these to make a DR curve. Here is an example I did years ago using a Canon DSLR:

 

Also, it is been mentioned already but log profiles do not introduce noise. They raise the blacks making the noise visible. If you see more noise in the graded image, you are not exposing the profile properly. Most log profiles need to be exposed slower than nominal because the nominal ISO of the profile is chosen to maximize some notion of  SNR, which is not necessarily the rating you want for a clean image (after grading). In fact, cameras do it all the time. Take a Sony A7 series camera, for example, and look through the minimal ISOs of their video profiles. See how the minimal ISO moves around between 100 and 3200 depending on the profile? That's because the camera is rated differently, depending on how it is supposed to redistribute tones according to the profile.

[maxotics]

5 hours ago, Bioskop.Inc said:

You can, and considering the topic at hand, should add a colour chart to the test - but if you do this White Balance the camera to the white piece of card (with BM cameras you'll need to test the various WB presets). You should also try to make sure the Black card is as black as possible - "The black card should be as black as possible, some people create a recessed box lined in black velvet".

First, Bioskop and CPC thanks for joining the conversation.  Because of the way LOG extends ISO I've never doubted for a second that there's be more noise in the image.  "add[ing] a colour chart to the test" is exactly what I've been trying to do.  First I used one-shot charts where I noticed that aggressive LOG recording gammas reduced the pre-grading color palette by a very large amount.  The gentler LOGs are another story.  Some people in the thread believe the LOG method of capturing color improves perceptual image quality, even if there is data loss compared to the "linear" profile.  In the end, they may be right!  But I want to see for myself through some tests

In order to test the color issue in more depth (AGAIN THANK YOU BIOSKOP for recognizing this should be a TO DO for anyone interested in this subject) I set up to compare how different gamma settings will capture all 16 million colors (assuming the camera is 8-bit).  First, I sliced apart full RGB images which you can get here https://allrgb.com/.  Here's a comparison of capturing part of this image on a monitor in EOS Cinema on the C100 (letting PS auto grad)

Now with EOS standard

If I'm just using my eye, I can't see much "artistic" difference between EOS std and Cinema.  So I can see how, in the "real world", filmmakers like @Oliver Daniel find this subject too technical because the differences aren't great enough to effect the final product.

I then wanted more control over the colors.  So wrote some Python/PIL scripts to generate 32 1920x1080 frame of all colors (4 pixels a color) where the colors are sorted by luminosity because what I really want to know is where in the DR LOG shooting plays games, so to speak.

Unfortunately, the aliasing issues of shooting RGB screens with RGB sensors have set me back.  There are too many artificial colors and noise for me to get at what I want.  Here's a graph of the screens and shooting with the A6300 in PP1 and PP5 

The original BMP of 32,410 colors.  32 frames of these are captured, 5 seconds a piece, by the A6300

PP1

PP5

The data suggests that PP5, or Cine 1, is capturing more color than Sony's PP1, which is pretty close to its standard profile (I believe).  So is @cantsin correct?  Is there an improvement of color using a LOG profile?  I still don't know.  Because both captures are recording more colors than the screen is meaning to display (it should create only 32,401 colors in our eye), due to re-mixing of RGB values, I can't tell which colors each gamma recording is getting from the proper RGB mixes, and which are essentially false aliasing colors.

To my eye, Cine 1 is loosing color fidelity, but I can't prove it with the data I have.  

To continue, I will need to created larger, probably 1-inch blocks of color on the monitor and average down to a hopefully noise-insignificant value. I'll also need to program stuff to put that data in a database and compare displayed color to captured. Don't know if I'm going to continue here.  That method will need hours of capture time (to go through the screens of color)  The really shrewd here might know what I could do, IF I could do this, so that is some motivation.  We'll just have to see.  

[Deadcode]

25 minutes ago, maxotics said:

 

 

Where do these lines shift if you increase the saturation or you increase the contrast with knee/black gamma settings? (but every other settings remains the same)

[maxotics]

4 minutes ago, Deadcode said:

Where do these lines shift if you increase the saturation or you increase the contrast with knee/black gamma settings? (but every other settings remains the same)

That exactly the kind of question I want to answer. Unfortunately, this data is so noisy I can't.  If I can eventually get good data then it would be wicked cool (at least to me) to see exactly how changes in the knee/black gamma settings change the tonality distribution.