Jump to content

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


maxotics
 Share

Recommended Posts

I did a video on this (2nd try) but it's very dull (link below).  My question to the forum is this, what am I missing in this logic?

  1. 8-bit video captures roughly 16 million color values
  2. Human vision is around 12 million (but let's call it 16 million)
  3. Human vision, is around 5-7 stops of dynamic range (without pupil change)
  4. CONCLUSION ONE: 8-bit color can deliver a complete color representation to a human; that is 16 million colors over a 6-stop gamut, let's say.  
  5. In S-LOG, one is widening the dynamic range to 10 stops, let's say, and spreading color information across the fixed 8-bit data space, which means we're losing saturation compared to the 16 million colors over 6 stops?
  6. Compounding the above's theoretical question, a sensor becomes noisy or erratic a few stops above and below a range where is can accurately do #4;  Therefore, doesn't one trade 16 million colors of better saturation for few colors and noise? (that's my finding after doing some experiments).

My current conclusion is S-LOG is not really about dynamic range, unless you interpret that as capturing only brightness values in the wild.  S-LOG is a "look".  I'm personally tired of it, but that has no bearing here.  The question is, isn't it a misnomer to say LUTs put the color back in?  When one shoots S-LOG, aren't they walking away with more noise than accurate color values in which to apply the LUT?  Will S-LOG look very dated and washed out in the future?  Why isn't there more discussion of the destructiveness of S-LOGs on color data?

Thoughts?

 

 

 

Link to comment
Share on other sites

1 hour ago, maxotics said:

I did a video on this (2nd try) but it's very dull (link below).  My question to the forum is this, what am I missing in this logic?

  1. 8-bit video captures roughly 16 million color values
  2. Human vision is around 12 million (but let's call it 16 million)
  3. Human vision, is around 5-7 stops of dynamic range (without pupil change)
  4. CONCLUSION ONE: 8-bit color can deliver a complete color representation to a human; that is 16 million colors over a 6-stop gamut, let's say.  
  5. In S-LOG, one is widening the dynamic range to 10 stops, let's say, and spreading color information across the fixed 8-bit data space, which means we're losing saturation compared to the 16 million colors over 6 stops?
  6. Compounding the above's theoretical question, a sensor becomes noisy or erratic a few stops above and below a range where is can accurately do #4;  Therefore, doesn't one trade 16 million colors of better saturation for few colors and noise? (that's my finding after doing some experiments).

My current conclusion is S-LOG is not really about dynamic range, unless you interpret that as capturing only brightness values in the wild.  S-LOG is a "look".  I'm personally tired of it, but that has no bearing here.  The question is, isn't it a misnomer to say LUTs put the color back in?  When one shoots S-LOG, aren't they walking away with more noise than accurate color values in which to apply the LUT?  Will S-LOG look very dated and washed out in the future?  Why isn't there more discussion of the destructiveness of S-LOGs on color data?

Thoughts?

 

S-LOGs are designed to compress the sensor's dynamic range in to a compressed codec to save disk space compared to raw.

Slog supposed to be recorded in 10 or 12 bit and it's designed to be colorgraded in post.

Colorgrading is much more than a simple "put a lut on it". If you record SLOG2 because someone said it's good, then put a REC709 lut on it you are wasting precious color information like you said. But if you do proper colorgrading it's much more important to have the scene's full dynamic range with full tonality (in 10bit)

I would not call SLOG3 "dated". It's very close to the HDR standard, and if i remember correctly you can make SLOG3 footage to HDR with a simple LUT.

Link to comment
Share on other sites

11 minutes ago, Deadcode said:

S-LOGs are designed to compress the sensor's dynamic range in to a compressed codec to save disk space compared to raw.

Slog supposed to be recorded in 10 or 12 bit and it's designed to be colorgraded in post.

Colorgrading is much more than a simple "put a lut on it". If you record SLOG2 because someone said it's good, then put a REC709 lut on it you are waisting precious color information like you said. But if you do proper colorgrading it's much more important to have the scene's full dynamic range with full tonality (in 10bit)

I would not call SLOG3 "dated". It's very close to the HDR standard, and if i remember correctly you can make SLOG3 footage to HDR with a simple LUT.

Sorry, let me clarify.  First my understanding is S-LOGs are designed to capture a wider dynamic range of scene brightness outdoors.  It is understood by the cinematographer that they will lose color fidelity.  Although image data can be compressed, is there such a thing as a "compressed CODEC"?  That is, whether shooting 8, 10, 12 bit H.264 or 14-bit channel RAW, you are limited by your largest data "word" value.  This is where I scratch my head.  Many people talk about it in a way that I think either I'm missing something or there is a lot of misconception out there about what CODECs do.  That is, no amount of compression can change the number of color values possible in the data space.

I wouldn't put it that REC709 wastes precious color information on SLOG, isn't it that SLOG wastes precious color information when it ends up in REC709, due to 1) sensor limitations and 2) the fixed size of your color space (8,10, whatever).  That's my question.

Link to comment
Share on other sites

2 minutes ago, maxotics said:

Sorry, let me clarify.  First my understanding is S-LOGs are designed to capture a wider dynamic range of scene brightness outdoors.  It is understood by the cinematographer that they will lose color fidelity.  Although image data can be compressed, is there such a thing as a "compressed CODEC"?  That is, whether shooting 8, 10, 12 bit H.264 or 14-bit channel RAW, you are limited by your largest data "word" value.  This is where I scratch my head.  Many people talk about it in a way that I think either I'm missing something or there is a lot of misconception out there about what CODECs do.  That is, no amount of compression can change the number of color values possible in the data space.

I wouldn't put it that REC709 wastes precious color information on SLOG, isn't it that SLOG wastes precious color information when it ends up in REC709.  That's my question.

You are correct. SLOG wastes color resolution if it end up in rec709  container. SLOG2 S-gamut is designed to emulate wider color gamut then Adobe RGB, but it's compressed to REC709's color gamut. So if you are using a LUT to recover the full gamut there will be "gaps" if the recording was 8 bit. (so most of the consumer Sony cameras doing this)

If you are recording 10 bit SLOG2 S-gamut it's not limited to 16,7 million colors and we no longer talking about the limits of rec709

If our goal is to represent the final footage on a rec709 display we have much more room to shift the colors and luminance in our footage the way we want. 

Most of the time you cant notice the difference in compression just the noise levels are little higher. So i think it's worth to use SLOGs but not at all costs and not in all situations (i mean in 8 bit, if i could record 10 bit, i would record slog all the time)

Link to comment
Share on other sites

S-log simply is a way to stuff more DR into 8-bit and 10-bit which both - with linear brightness values - can't hold 14 or 15 stops. Of course if you map the values to your liking to rec_709, practically all colors and nuances are just interpolations. Therefore, compared to raw, which just provides quantitatively more information, you will introduce visible artifacts sooner. That's the difference. S-log is sometimes called the poor man's raw. 

Link to comment
Share on other sites

I believe I know what you're saying when you say "compressed to REC709's color gamut".  But isn't compressed a bad word?  Compressed usually means you're doing a bunch of math to represent a complex image, say in a smaller data file because a lot of the data is repeated.  That is, it's easy to imagine how compression could remove a lot of data from an image of the ocean and sky, but not an image of a fine tapestry.  Isn't it more like truncated?  No matter what kind of LUT you use, there is missing color information?  

I'm glad you brought up 10-bit SLOG.  I don't see much of a difference between 8-bit SLOG and 10-bit SLOG (on Sony cameras) because the 9th and 10th bit are used AFTER the sensor has been read.  That is, I believe many people believe the higher bit resolution approaches the 14-bits of single-channel RAW, when it remains apples to oranges comparison.  I haven't done any tests using 10-bit, but I suspect I would still discover nearly 30% color lost in an aggressive S-LOG, as I found in 8-bit.

13 minutes ago, Axel said:

S-log simply is a way to stuff more DR into 8-bit and 10-bit which both - with linear brightness values - can't hold 14 or 15 stops. Of course if you map the values to your liking to rec_709, practically all colors and nuances are just interpolations. Therefore, compared to raw, which just provides quantitatively more information, you will introduce visible artifacts sooner. That's the difference. S-log is sometimes called the poor man's raw. 

Is there anything remotely simple about S-LOG ;)  Let me explain a different way.  Your monitor can display 16 million colors.  If it's 1080 it'll only show 2 million, so you need about 8 images to show the full capability of your display.  If you were taking a video of 8 drawings you created on your computer, and the 8 drawings all added up to 16 million colors you'd want your camera to record every color, right? If you shoot a normal profile and look at each pixel of color in your video you will count up 16 million.  If you shoot S-LOG, you will get, say 10 million. 

Link to comment
Share on other sites

7 minutes ago, maxotics said:

I believe I know what you're saying when you say "compressed to REC709's color gamut".  But isn't compressed a bad word?  Compressed usually means you're doing a bunch of math to represent a complex image, say in a smaller data file because a lot of the data is repeated.  That is, it's easy to imagine how compression could remove a lot of data from an image of the ocean and sky, but not an image of a fine tapestry.  Isn't it more like truncated?  No matter what kind of LUT you use, there is missing color information?  

I'm glad you brought up 10-bit SLOG.  I don't see much of a difference between 8-bit SLOG and 10-bit SLOG (on Sony cameras) because the 9th and 10th bit are used AFTER the sensor has been read.  That is, I believe many people believe the higher bit resolution approaches the 14-bits of single-channel RAW, when it remains apples to oranges comparison.  I haven't done any tests using 10-bit, but I suspect I would still discover nearly 30% color lost in an aggressive S-LOG, as I found in 8-bit.

S-LOG is totally worth it, if you now how to use and treat it. 

It's not only the extended dynamic range, its the impact it has on highlight rolloff which makes the transitions from shadows to mids to highlights very smooth. The difference is night and day. 

If you know what to do, using S-LOG will give you a far superior look to standard profiles. 

I personally modify SLOG 2 for better colour and I love it. :)

Link to comment
Share on other sites

I'm not sure if you're talking a gamma curve or color gammut, because you keep going back and forth describing luma and chroma. SLog only affects the recorded luma values, though it's designed to be used with it's corresponding color gammut (SLog2 with S.Gammut, SLog3 with S.Gammut3, etc.)... Not trying to be overly technical or anything but you said you may be missing something and maybe this is it.

 

I can say from experience that there are certainly trade-offs to trying to using these, such as lost detail and increased noise, but ultimately it gives you much more control in post over how the final image will look. One caveat to this is that 8 bit 4:2:0 is not a great place to try to store SLog (esp. 3) and wide color gammuts. 10 bit 4:2:2 and higher are MUCH better for this.

14 minutes ago, maxotics said:

I haven't done any tests using 10-bit, but I suspect I would still discover nearly 30% color lost in an aggressive S-LOG, as I found in 8-bit.

This is telling. You've just got to experience it: 10 bit 4:2:2 is night and day different than 8 bit 4:2:0 when it comes to color grading.

Link to comment
Share on other sites

Just now, Oliver Daniel said:

S-LOG is totally worth it, if you now how to use and treat it. 

It's not only the extended dynamic range, its the impact it has on highlight rolloff which makes the transitions from shadows to mids to highlights very smooth. The difference is night and day. 

If you know what to do, using S-LOG will give you a far superior look to standard profiles. 

I personally modify SLOG 2 for better colour and I love it. :)

Oliver, you know I love you to death and greatly admire your work.  But I have to push back a bit here.  When you say "S-LOG is totally worth it, if you now now to use and treat it." You're insinuating that SLOG is better than standard profiles, that everyone should shoot it?  Right?  It is ONLY extended dynamic range in the fact that it picks up extended brightness values in the scene, it is NOT extended COLOR range, in fact, quite the opposite.  The extended brightness range you get are TRADED OFF for color fidelity.  Now, you may only want to see those 10 million colors, and I respect that, but to say that everyone should shoot in a way that degrades the camera FULL color fidelity I don't understand.  

As for highlight rolloff, yes, I can see that, but again, it ignores what you lose in the mid-tones.

Please think about what I wrote more carefully and tell me why S-LOG does not trade off "brightness contrast" in a scene for color fidelity in the full gamut we expect.  Tell me why what you're saying is an objective statement that SLOG is better than standard profiles.

Link to comment
Share on other sites

24 minutes ago, maxotics said:

I'm glad you brought up 10-bit SLOG.  I don't see much of a difference between 8-bit SLOG and 10-bit SLOG (on Sony cameras) because the 9th and 10th bit are used AFTER the sensor has been read.  That is, I believe many people believe the higher bit resolution approaches the 14-bits of single-channel RAW, when it remains apples to oranges comparison.  I haven't done any tests using 10-bit, but I suspect I would still discover nearly 30% color lost in an aggressive S-LOG, as I found in 8-bit.

I

if your goal is to grade for rec709 display you are not loosing any color informations, because 10bit slog contains all of what can be represented on a rec709 display.

And yes, probably not the "compression" is the best word here

13 minutes ago, EthanAlexander said:

I'm not sure if you're talking a gamma curve or color gammut, because you keep going back and forth describing luma and chroma. SLog only affects the recorded luma values, though it's designed to be used with it's corresponding color gammut (SLog2 with S.Gammut, SLog3 with S.Gammut3, etc.)... Not trying to be overly technical or anything but you said you may be missing something and maybe this is it.

 

I can say from experience that there are certainly trade-offs to trying to using these, such as lost detail and increased noise, but ultimately it gives you much more control in post over how the final image will look. One caveat to this is that 8 bit 4:2:0 is not a great place to try to store SLog (esp. 3) and wide color gammuts. 10 bit 4:2:2 and higher are MUCH better for this.

This is a good representation in Andrew's EosHD Pro color what happens when it needs wider color gamut than REC709. With S-gamut yes there are fewer steps but much wider range of color can be represented.

 

 

Link to comment
Share on other sites

1 minute ago, Deadcode said:

if your goal is to grade for rec709 display you are not loosing any color informations, because 10bit slog contains all of what can be represented on a rec709 display.

And yes, probably not the "compression" is the best word here

This is a good representation in Andrew's EosHD Pro color what happens when it needs wider color gamut than REC709. With S-gamut yes there are fewer steps but much wider range of color can be represented.

 

I see that 10bit is better than 8bit, certainly, only pointing out that it doesn't bring one closer to RAW in color depth (in my experience).    Can we see a wider color gamut than REC709?  We can certainly see one in nature because things can be brighter than 6 stops, say, and our pupils dilate, but in the real world, our displays and needs don't exceed it?

I want to make a point again, that you can get beautiful footage from S-LOG, I'm not talking about subjective benefits of S-LOG.  Or to put it another way, if all cameras started with S-LOG and today they introduced standard profiles everyone would be like "Wow, look at all those rich colors!" :)

Link to comment
Share on other sites

19 minutes ago, EthanAlexander said:

I still don't understand why you're jumping back and forth between stops (luma values) and and color gammut. Gamma curves affect perception of saturation but they have absolutely nothing to do with the color space that is recorded.

How do they have nothing to do with color space recorded?  What is your definition of color?  Gamut is applied to color values after they are recorded.  Tell me if I'm missing something.  So gamut has nothing to do with the amount of colors you record in an 8bit space.  Or it has everything to do with it because when you record into an extended gamut what happens?

@EthanAlexander, don't get me wrong.  This was extremely confusing to me for a long time!  Maybe still is!  What I didn't understand is although the camera records brighness, it doesn't not end up as brightness in video data, it only ends up as colors which have a gamut applied to them later to give them brightness.

Link to comment
Share on other sites

9 minutes ago, EthanAlexander said:

I still don't understand why you're jumping back and forth between stops (luma values) and and color gammut. Gamma curves affect perception of saturation but they have absolutely nothing to do with the color space that is recorded.

I think in 8 bit recording you have certain amount of color represented including max saturation. With wider color gamut you have more visible colors (for example when viewing raw image on Adobe RGB display) and more saturated colors possible. If you want to represent wider color gamut then rec709 in a rec709 container you loose tonality. Which means if there is a green apple on the table and the left side of that apple is "more" green the the right side, with this "compression" the apple's saturation will be the same for the whole apple after the recovery.

But this is just my understanding from this awesome article

Link to comment
Share on other sites

44 minutes ago, maxotics said:

How do they have nothing to do with color space recorded?  What is your definition of color?  Gamut is applied to color values after they are recorded.  Tell me if I'm missing something.  So gamut has nothing to do with the amount of colors you record in an 8bit space.  Or it has everything to do with it because when you record into an extended gamut what happens?

Ok, I think I understand the missing part: 8 bit does have a limitation on the number of possible colors recorded (as does 10, 12, 14 bit etc), but we have a choice over WHICH colors we are trying to fit in to that fixed number of colors.

So, for instance, with REC709, sRGB has become the defacto color space online because it leads to a high perception of saturation and for the most part looks very true to life while not requiring expensive displays to see them. However, by choosing our gamut, we are telling the camera which subset of "real" colors we want to try to record in the limited space we have (in the case of 8 bit: 16M colors). So, in practice, this is why on a Sony camera we choose both a gamma curve (which affects how luma values are treated) in the form of S Log2, S Log 3, etc. and also have the choice over color gamut, such as standard, pro, cinema, S.Gamut, S.Gamut3, and S.Gamut3/Cine. These color spaces dictate what subset of the spectrum we're trying to squeeze into our codec.

Hopefully that clears some things up. Alternatively, maybe I'm missing your greater point.

Link to comment
Share on other sites

19 minutes ago, EthanAlexander said:

Ok, I think I understand the missing part: 8 bit does have a limitation on the number of possible colors recorded (as does 10, 12, 14 bit etc), but we have a choice over WHICH colors we are trying to fit in to that fixed number of colors.

So, for instance, with REC709, sRGB has become the defacto color space online because it leads to a high perception of saturation and for the most part looks very true to life while not requiring expensive displays to see them. However, by choosing our gammut, we are telling the camera which subset of "real" colors we want to try to record in the limited space we have (in the case of 8 bit: 16M colors). So, in practice, this is why on a Sony camera we choose both a gamma curve (which affects how luma values are treated) in the form of S Log2, S Log 3, etc. and also have the choice over color gammut, such as standard, pro, cinema, S.Gammut, S.Gammut3, and S.Gammut3/Cine. These color spaces dictate what subset of the spectrum we're trying to squeeze into our codec.

Hopefully that clears some things up. Alternatively, maybe I'm missing your greater point.

Let me try again.  You're in a well lit room where you can shoot 100 ISO (or whatever allows the camera to get the best saturation of colors--lowest noise).   The brightest part of the room is 6 stops greater than the dimmest part.  There are 16 million colors which one can discern on the objects.  The camera records 256 values from each pixel, either a R,G,B, and put them together as a 24-bit value, 1 to 16 million.  These values combine both color information (what filter is over the pixel) and brightness information.  However, they are just data points and in our 24-bit full color space we can fit them all in, but no more.  When we view the image, our display superimposes a GAMUT over these values so the 1 shade of RED say is almost white, as bright as it can be, and the 255 is almost black, or as dark as the red can be.  In a sense, we have 2.6 million colors per stop of gamut.  If we extend our GAMUT on our display, which is possible on some high end models, by even 1 stop, then we are not getting any more color information because, again, we only have 16 million.  It will look different, we may like that look, but it isn't actually giving us more color-wise, or in hues?

When you shoot S-LOG you're using a GAMUT that is not really visible to us in real life, that's my understanding now.  Because you always have to bring that gamut down to 6 stops for the reality of our equipment (and viewing preference).  When we talk about "compressing" that information down, we're not recognizing that when we record into that extended gamut in 8 bit, there are lot of color holes.  Those holes are empty for good.  A LUT can't replace a color that isn't there with a better color, no matter how nice that color might be ;)

 

 

Link to comment
Share on other sites

46 minutes ago, maxotics said:

How do they have nothing to do with color space recorded?  What is your definition of color?  Gamut is applied to color values after they are recorded.  Tell me if I'm missing something.  So gamut has nothing to do with the amount of colors you record in an 8bit space.  Or it has everything to do with it because when you record into an extended gamut what happens?, don't get me wrong.  This was extremely confusing to me for a long time!  Maybe still is!  What I didn't understand is although the camera records brighness, it doesn't not end up as brightness in video data, it only ends up as colors which have a gamut applied to them later to give them brightness.

In context to my previous response, I think this has more to do with our perception of saturation than the actual colors that are being recorded. For instance, I know what by increasing the "saturation slider" in photoshop, it's actually throwing away a lot of colors and narrowing the overall number still left in the image, because of how our eyes and brains work.

Honestly, don't get the full science of it all and never will, I just know from experience that choosing a gamma curve and color space separately will lead to different results.

8 minutes ago, maxotics said:

Let me try again.  You're in a well lit room where you can shoot 100 ISO (or whatever allows the camera to get the best saturation of colors--lowest noise).   The brightest part of the room is 6 stops greater than the dimmest part.  There are 16 million colors which one can discern on the objects.  The camera records 256 values from each pixel, either a R,G,B, and put them together as a 24-bit value, 1 to 16 million.  These values combine both color information (what filter is over the pixel) and brightness information.  However, they are just data points and in our 24-bit full color space we can fit them all in, but no more.  When we view the image, our display superimposes a GAMUT over these values so the 1 shade of RED say is almost white, as bright as it can be, and the 255 is almost black, or as dark as the red can be.  In a sense, we have 2.6 million colors per stop of gamut.  If we extend our GAMUT on our display, which is possible on some high end models, by even 1 stop, then we are not getting any more color information because, again, we only have 16 million.  It will look different, we may like that look, but it isn't actually giving us more color-wise, or in hues?

When you shoot S-LOG you're using a GAMUT that is not really visible to us in real life, that's my understanding now.  Because you always have to bring that gamut down to 6 stops for the reality of our equipment (and viewing preference).  When we talk about "compressing" that information down, we're not recognizing that when we record into that extended gamut in 8 bit, there are lot of color holes.  Those holes are empty for good.  A LUT can't replace a color that isn't there with a better color, no matter how nice that color might be ;)

 

 

I really don't know if I'm not understanding or you're not. We've reached the extent of my knowledge ;)

Link to comment
Share on other sites

i have a question which is just partly related to the original topic.

If i record ML raw with my 5D2, am i able to create proper HDR footage out of it? i know it's limited to 11.7 stop DR and HDR requires 15 EV, but HDR is just a tonemapping right?

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...