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

[maxotics]

1 hour ago, EthanAlexander said:

"So designing a log curve that is optimized for the camera's sensor and its compression processing will generate superior shadow quality through compression processing than will linear compression, but without visible impact to the highlights." https://cineform.blogspot.com/2007/09/10-bit-log-vs-12-bit-linear.html @maxotics (as posted by @iaremrsir)

One of the posts recommended in that article really nails a lot of issues (and I learned something too!). https://prolost.com/blog/rawvslog  That guy knows his stuff.  However, as I understand it, there is a big difference between the kind of LOG you get on a consumer camera, and the kind of LOG he's talking about, which is how LOG can reduce RAW data through a type of perceptual compression.  In Magic Lantern RAW, for example, you get each pixel's value in a 14-bit single channel space.  What he's saying is the distance between those values numerically, does not match the distance we perceive.  Therefore, you can re-map the values onto a LOG curve and retain the same amount of information.  

I don't see a similarity in consumer LOG to the RAW LOG they're talking about in those articles.  Maybe I'm missing something.

3 hours ago, AndrewM said:

I don't think I understand the OP. We are all in option 3, right? And Log is a tool, designed to make the best of certain situations. If we are in option 1, then no log. If the costs of log in a particular situation outweigh the benefits, then don't use it. If you are just using log as a kind of "bracketing" to avoid not having your shot, then that is fine (but you can learn to do better in some situations, and you have to recognize the costs). If you like the "log look" and don't grade, then... that is your preference.

Are we "all in option 3"?  I don't believe we are.  I believe many people are confused about how LOG works and believe it to be a superior profile to shoot everything.  I believe this because people who should know better do tests on YouTube where they're shooting LOG in low DR scenes.  Your Option 1.  Do you really believe I am making an argument that LOG is useless?  Please   I've written so much that for you to say you don't understand my questions is a bit disheartening.

My question is yours, "If the costs of log in a particular situation outweigh the benefits, then don't use it."  What are the costs?  Can you point me to objective guidance that doesn't resort to subjective feel?  

 

[cantsin]

25 minutes ago, maxotics said:

One of the posts recommended in that article really nails a lot of issues (and I learned something too!). https://prolost.com/blog/rawvslog  That guy knows his stuff.  However, as I understand it, there is a big difference between the kind of LOG you get on a consumer camera, and the kind of LOG he's talking about, which is how LOG can reduce RAW data through a type of perceptual compression.  In Magic Lantern RAW, for example, you get each pixel's value in a 14-bit single channel space.  What he's saying is the distance between those values numerically, does not match the distance we perceive.  Therefore, you can re-map the values onto a LOG curve and retain the same amount of information.  

I don't see a similarity in consumer LOG to the RAW LOG they're talking about in those articles.  Maybe I'm missing something.

"Consumer Log" is the same thing since it's also log created from the raw sensor data. In the case of Sony S-Log and Panasonic S-Log, it also has a much bigger color space than Rec709/RGB. The real issue of "consumer log" is that when this signal, which requires high bandwidth encoding, is compressed into a small bandwidth 8bit 4:2:0 codec (as was the case with the Panasonic GH4 and still is the case with Sony's RX/a6x00/a7x cameras).

Again, color resolution behaves equivalent to pixel resolution. To stay in this analogy: If you compress a 48MP picture (shot with an A7r) with 10% JPEG quality in order not to exceed 1 MB file size, you end up with an image that still has 48MP but looks terrible because of the extreme compression artefacts (see below; the original image is here: https://4.img-dpreview.com/files/p/TS7952x5304~sample_galleries/7076458101/9405098100.jpg). In such a case, it's better to shoot or store with a smaller pixel resolution but higher jpeg quality. The equivalent happens with color if a high dynamic range log image is stored with "consumer" bit depths and "consumer" data compression ratios - i.e. in a bit depth and codec that isn't sufficient for high-end, studio quality mastering, and completely inadequate for an extended color space such as s-gamut. In such a case, it's better to shoot Rec709.

 

[maxotics]

25 minutes ago, cantsin said:

"Consumer Log" is the same thing since it's also log created from the raw sensor data. In the case of Sony S-Log and Panasonic S-Log, it also has a much bigger color space than Rec709/RGB. The real issue of "consumer log" is that when this signal, which requires high bandwidth encoding, is compressed into a small bandwidth 8bit 4:2:0 codec (as was the case with the Panasonic GH4 and still is the case with Sony's RX/a6x00/a7x cameras).
 

Okay, so I should see a huge difference in the difference of color capture between the A6300 or C100, internal 8bit 4:2:0 video vs HDMI to ProRes HQ?  What do you predict it will be?  That is, the percentage difference in color details.  So if the internal captures 50 colors and the HDMI 100 colors, then a 100% difference.

BTW, I understand this is the point Eddie is making.  My question has always been a simple one between LOG and standard profile in any camera, so not looking at the theoretical difference IF the camera had 4:4:4 compression,say.  However, happy to explore that!

[cantsin]

27 minutes ago, maxotics said:

Okay, so I should see a huge difference in the difference of color capture between the A6300 or C100, internal 8bit 4:2:0 video vs HDMI to ProRes HQ?

The comparison doesn't really help because (a) both cameras only output 8bit via HDMI, (b) they have completely different log curves. Slog-2/3 was developed for Sony's top-of-the line cinema cameras (such as the F5), optimized for their sensors and codecs, and is implemented in the consumer cameras only as a compatibility function. (So that F5 operators can use an a6300 or A7s as a c-cam or crash cam and shoot footage that perfectly mixes with F5 footage. Same for V-Log as an option for using GH4/5 footage as c-cam footage in projects shot with Varicams.) 

Canon's CLog (not to be mixed up with the newer Canon Log....) is an entirely different beast because it was developed for the first generation of C100/C300 cameras and is optimized for their sensor and 8bit 4:2:2 codec. This also explains why it is a much less aggressive log curve than SLog2/3.

[maxotics]

Hi @cantsin don't get mad at me but you're doing exactly what drives me crazy.  The difference between LOG, or any other image IQ, between an F5, with 9 Megapixels (made only to shoot video), vs a Sony A6300 with 24 Megapixels (and made to take photographs) has more to do with the physics of the sensors than optimization and CODECs.  To say that LOG is "implemented in the consumer cameras only as a compatibility function" implies that they COULD have matched the F5 if they wanted to, when I'd call it impossible due to the different sensor types.  Yes, no?  I don't know why you're bringing F5s into this question?  Sorry!

Here are four screen captures from a Canon C100.  These are the color counts for each

           71,685 cinema.png

         227,536 eos_std.png

         162,560 no prfile.png

         148,903 wide_dr.png

So if you shoot "wide dr", you're only getting 65% of the tonality you'd get from the standard profile.  I understand that in the real world, grading can make a pleasing image out of all these profiles.  

One question everyone seems to aggressively ignore, is HOW MUCH would a wide dr profile hurt a scene shot within 5 stops of dynamic range?  

 

[cantsin]

31 minutes ago, maxotics said:

Hi @cantsin don't get mad at me but you're doing exactly what drives me crazy.  The difference between LOG, or any other image IQ, between an F5, with 9 Megapixels (made only to shoot video), vs a Sony A6300 with 24 Megapixels (and made to take photographs) has more to do with the physics of the sensors than optimization and CODECs.  To say that LOG is "implemented in the consumer cameras only as a compatibility function" implies that they COULD have matched the F5 if they wanted to, when I'd call it impossible due to the different sensor types.  Yes, no?  I don't know why you're bringing F5s into this question?  Sorry!

No, I did not mean to imply this at all. Sorry if it came across that way (but this is an easy misunderstanding on a forum such as EOSHD). What I wanted to get across is this is:

- a Log function normally is made to measure for a particular camera sensor, its dynamic range and often also the bandwidth of the camera's codec. (A good example is Blackmagic which uses different Logs, and different corresponding LUTs for Rec709 conversion, for its different camera families.)

- sLog2 and sLog3 were engineered for Sony's professional cinema cameras with their high dynamic range and studio-grade XAVC 10bit 400-600Mbit/s codecs.

- sLog2 and sLog3 are also implemented in Sony's 8bit consumer cameras never mind the fact that these log curves are neither optimized for their sensors (which likely have 1-2 stops less usable dynamic range), nor meant to be stored in low-bitrate 8bit 4:2:0. 

- Even if Sony's consumer cameras had better codecs AND a 10bit pipeline (which would require a complete reengineering of the cameras, and likely wouldn't be feasible with today's technology when keeping the same compact body sizes - given the fact that Sony's cameras already have overheating issues with their current feature set), sLog2 and sLog3 wouldn't be the perfect log curves for them because, as stated, they aren't optimized for the sensors of these cameras. The same is true for the GH4/GH5: Since V-Log has been engineered for the Varicam with its higher dynamic range, the German site Slashcam found out that when using it in the GH4 and GH5, the upper bit in the recording (which stores the highest exposed part of the image) remains unused, simply because the GH4/GH5 sensor clips at that exposure. This effectively reduces the 10bit recording of the GH5 (or GH4 with an external recording) to merely 9bit in V-Log, the price you pay for Varicam compatibility.

- The advantage of using Log curves engineered for higher-end cameras in lower-end (consumer) cameras is that the footage of both camera types can be easily mixed in the same timeline and graded with the same LUTs. So, in this sense, SLog2/3 and Vlog aren't engineered for maxing-out the video quality potential of cameras like the A6300 or the GH5, but to make them easier to use as c-cams/crashcams in projects where the larger cinema cameras are being used. 

- So, in the best case for DIY/low-end shooters, we would have Log curves engineered for "our" cameras, their sensor and the dynamic range of those sensors - instead of "alien" log curves that were engineered for different, higher-end cameras.

- The only exception so far (aside from Blackmagic) has been Canon with its CLog curve for the original, 8bit C100/C300. It's a "light" log curve engineered to preserve the dynamic range of the sensor in these cameras and not to be destroyed/yield questionable results in 8bit recording.

 

[maxotics]

16 minutes ago, cantsin said:

No, I did not mean to imply this at all. Sorry if it came across that way (but this is an easy misunderstanding on a forum such as EOSHD). What I wanted to get across is this is:...

I'm sorry too.  I'm learning a lot from you.  THANKS!  It just tweaks me when people (not you) explain the difference between different sandboxes, buckets, whatever, as if improvements can't be made in deciding how to chose a LOG profile or not.  I want to find metrics that are more actionable.  As I say in the beginning, my feeling is there is too much hype and mis-understandings surrounding these LOG profiles that are hurting filmmakers.  I'm trying to figure out scientific tests to explore that further.

[EthanAlexander]

1 hour ago, maxotics said:

Here are four screen captures from a Canon C100.

You're shooting something on a monitor that already has it's own limited dynamic range and color representation, so I don't see how this is helpful... Am I wrong that this would be biased?

[mercer]

Sorry to jump in the mix here, but I have a somewhat relevant question that I've been curious about for a while...

I had the a6500 last year and was thoroughly impressed with its 4K IQ, using sLog3. It is the closest I've seen to Raw with an 8bit, consumer camera.

Obviously, the major issue with using sLog3 is the potential for banding and artifacts, but if one chooses monochrome for the color gamma, will the 4 luma values in its 4:20 codec help to combat the banding and artifacts, or since the gamma is down the pipeline, there can still be an issue?

And if there is a benefit with monochrome, does the 4 luma values have a positive IQ effect with all 8bit cameras when shooting, in camera, B&W or would it need to be a monochrome specific sensor like the one from Leica or Digital Bolex?

[EthanAlexander]

51 minutes ago, cantsin said:

- sLog2 and sLog3 are also implemented in Sony's 8bit consumer cameras never mind the fact that these log curves are neither optimized for their sensors (which likely have 1-2 stops less usable dynamic range), nor meant to be stored in low-bitrate 8bit 4:2:0.

I'm not sure you're correct that the sensors aren't able to handle SLog2/3... I think it's just that theses curves weren't designed for 8 bit 4:2:0 and low bit rates. For instance: on the FS5... 10 bit 4:2:2 1080p performs night-and-day better than 8 bit 4:2:0 4K, and it's the same sensor for each (obviously).

[maxotics]

11 minutes ago, EthanAlexander said:

You're shooting something on a monitor that already has it's own limited dynamic range and color representation, so I don't see how this is helpful... Am I wrong that this would be biased?

Thanks for the question. Yes, it is biased towards the display's DR.  My reasoning is that the viewer only experiences the world through a display (which is probably of less quality than mine).  Therefore, we should expect the camera to be able to record whatever is displayed on the monitor.  My test won't be to see how much of the monitor's limited dynamic range the camera can record (it will be less for a variety of reasons), but what the difference is in fidelity between the camera's default settings and a LOG profile.  

If that simple initial test holds true, if I shot both standard profile and Cinema in my office I should expect a 35% color loss from the Cinema profile.  Again, I understand what many have pointed out, that most people would NOT shoot LOG in my office.  One of my goals is to quantify the quality loss of LOG profiles in standard DR situations.  In order to do that, I need to capture the whole 24-bit color space, which so far, is going to be 32 screens of 1920x1080 colors, where each color will get 4 pixels.

Make sense?

[EthanAlexander]

3 minutes ago, mercer said:

Sorry to jump in the mix here, but I have a somewhat relevant question that I've been curious about for a while...

I had the a6500 last year and was thoroughly impressed with its 4K IQ, using sLog3. It is the closest I've seen to Raw with an 8bit, consumer camera.

Obviously, the major issue with using sLog3 is the potential for banding and artifacts, but if one chooses monochrome for the color gamma, will the 4 luma values in its 4:20 codec help to combat the banding and artifacts, or since the gamma is down the pipeline, there can still be an issue?

And if there is a benefit with monochrome, does the 4 luma values have a positive IQ effect with all 8bit cameras when shooting, in camera, B&W or would it need to be a monochrome specific sensor like the one from Leica or Digital Bolex?

Since banding is mostly due to color subsampling rather than 8 bit, I'd say yes, but it's worth testing. Can't you just take a sample you shot and remove the chroma to test?

[mercer]

1 minute ago, EthanAlexander said:

Since banding is mostly due to color subsampling rather than 8 bit, I'd say yes, but it's worth testing. Can't you just take a sample you shot and remove the chroma to test?

Honestly, when I went with the 5D3, I deleted a lot of my test footage from most of my past cameras for the storage space on my drives, but I'll check to see what I have. Good idea. For some reason I thought it would have to occur in camera to have a positive effect. Thanks. 

[cantsin]

9 minutes ago, EthanAlexander said:

I'm not sure you're correct that the sensors aren't able to handle SLog2/3... I think it's just that theses curves weren't designed for 8 bit 4:2:0 and low bit rates. For instance: on the FS5... 10 bit 4:2:2 1080p performs night-and-day better than 8 bit 4:2:0 4K, and it's the same sensor for each (obviously).

Well, it's not that a sensor cannot "handle" a log curve, but the question is whether a log curve can handle/fit the dynamic range of the sensor. If the log curve has been designed for the highest-end, highest dynamic range sensor, it won't be the optimal fit for a lower-end/lower dynamic range sensor. For example, if a lower-end sensor has 2 stops less usable dynamic range (note the word usable... on a paper, a sensor can have a lot of dynamic range, but if the two lowest stops are more noise than signal, those stops don't really count) and the log curve records those 2 stops in one bit, you'll be throwing away one bit of your color depth. In the worst case, you might end up recording effective 8bit in log even when the codec is 10bit.

[cantsin]

13 minutes ago, maxotics said:

Thanks for the question. Yes, it is biased towards the display's DR.  My reasoning is that the viewer only experiences the world through a display (which is probably of less quality than mine).  Therefore, we should expect the camera to be able to record whatever is displayed on the monitor.  My test won't be to see how much of the monitor's limited dynamic range the camera can record (it will be less for a variety of reasons), but what the difference is in fidelity between the camera's default settings and a LOG profile.  

If that simple initial test holds true, if I shot both standard profile and Cinema in my office I should expect a 35% color loss from the Cinema profile.  Again, I understand what many have pointed out, that most people would NOT shoot LOG in my office.  One of my goals is to quantify the quality loss of LOG profiles in standard DR situations.  In order to do that, I need to capture the whole 24-bit color space, which so far, is going to be 32 screens of 1920x1080 colors, where each color will get 4 pixels.

Make sense?

No, because this mixes up display dynamic range with scene dynamic range.

When we speak of the dynamic range a camera can record, we refer to scene dynamic range: For example, 14 stops from shadows to highlights in an outdoor daylight scene you record. A typical, good computer monitor has a contrast ratio of 1:1000, that's 10 stops (2^10=1024). What we do in video postproduction is to compress those 14 stops you recorded outdoors into the 10 stops of your monitor, or, in the case of 8bit Rec709/8bit RGB, the mere 8 stops supported by this display and color standard. In other words, we align the brightest stops of the motive with the brightest stops of the display, the lowest with the lowest, and the medium is where our gamma sits. So we'll roughly end up mapping stops 11-12 of the scene to stop 8 of the display, stops 9-10 to stop 7, stop 7-8 to stop 6, stop 6 to stop 5, stop 5 to 4 etc. all the way down to stop 1-2 to stop 1. That's exactly what a Log-to-Rec709 LUT will do. 

If the camera sensor dynamic range were identical to that of the display, or if we would only record within the same dynamic range as the scene, the highlights of the scene would clip and the shadows would be black.

- If I'm not mistaken, you're caught in math trap where you base all your assumptions on the fact that the human eye can see about 16 million colors and that an 8bit true color display displays 16 million colors, too. So why do we need more than those 16 million colors? Because this number doesn't take into account the adaptation of the human eye (in combination with the brain). While the human eye can physically only see about 7 stops, it adjusts its iris so quickly (and we usually scan a scene with our eyes going from shadows to highlight) that our brain effectively "sees" a higher dynamic range. 

[EthanAlexander]

3 hours ago, maxotics said:

One of the posts recommended in that article really nails a lot of issues (and I learned something too!). https://prolost.com/blog/rawvslog  That guy knows his stuff. 

Great find.

19 minutes ago, mercer said:

Honestly, when I went with the 5D3, I deleted a lot of my test footage from most of my past cameras for the storage space on my drives, but I'll check to see what I have. Good idea. For some reason I thought it would have to occur in camera to have a positive effect. Thanks. 

I owned an a6500 for like 3 weeks but never used it for a paid gig. I'll see if I have any footage I can dig up.

[maxotics]

12 minutes ago, cantsin said:

No, because this mixes up display dynamic range with scene dynamic range.

That's EXACTLY the confusion I'm trying to remove.  What is the camera's best shot at a perfect image in DR that does not make our pupils dilate?  

7 minutes ago, EthanAlexander said:

Great find.

Wouldn't have found it without you! Thanks!

[mercer]

18 minutes ago, EthanAlexander said:

Great find.

I owned an a6500 for like 3 weeks but never used it for a paid gig. I'll see if I have any footage I can dig up.

That's about how long I had mine. I was planning on keeping it but on the same day I found a bunch of 5D3 Raw videos, I got an overheating warning on the a6500 on a 40 degree, Fahrenheit day after shooting two 15 second clips in the shade. I knew my next project was going to be in the Summer and I didn't want to risk it. But for winter shooting, in B&W, the a6500 could be a beast of a camera for handheld, run and gun... especially if the price drops a little over the next few months.

Thanks, for checking your files.

[maxotics]

2 hours ago, mercer said:

Sorry to jump in the mix here, but I have a somewhat relevant question that I've been curious about for a while...

I had the a6500 last year and was thoroughly impressed with its 4K IQ, using sLog3. It is the closest I've seen to Raw with an 8bit, consumer camera.

Obviously, the major issue with using sLog3 is the potential for banding and artifacts, but if one chooses monochrome for the color gamma, will the 4 luma values in its 4:20 codec help to combat the banding and artifacts, or since the gamma is down the pipeline, there can still be an issue?

And if there is a benefit with monochrome, does the 4 luma values have a positive IQ effect with all 8bit cameras when shooting, in camera, B&W or would it need to be a monochrome specific sensor like the one from Leica or Digital Bolex?

Here are two images shot with an a6300 in 4K at a red gradient on my monitor.  One is PP1 the other PP7  This is a quick and dirty test where I shot the screen, curved/exposed in PS to match best I could, red to white.  I then saved as JPG at quality 1 (lowest) to bring out the banding issues.  In my opinion, it would be impossible to shoot LOG with the a6300/a6500 without getting banding issues if the scene has a subject where it would be noticeable.  My current guess is that PP1 in the a6300 has 500% more color than PP7.  I'll know more later.  

Anyway, I don't see how monochrome would have any improvements if the camera can't get a fully saturated "red" patch together

 

[cantsin]

14 hours ago, maxotics said:

That's EXACTLY the confusion I'm trying to remove.  What is the camera's best shot at a perfect image in DR that does not make our pupils dilate?  

But the simple truth is that you need extended dynamic range (and thus more than standard Rec709 capture) to record an image like this (where Rec709 may be the display color sapce):

...unless you use an incredible amount of reflectors and on-set lights as Sergio Leone's DP probably did for this scene. (In addition to using negative film stock with its much extended dynamic range over reversal/positive film. It also helps to think of Log video as an equivalent of negative stock and of Rec709 capture as the equivalent of reversal film.)

Today, we can capture such scenes without artificial light thanks to the high dynamic range of modern cameras. But if the above picture would have been shot with an in-camera Rec709 profile, the house front would be black, the sky would be white and Charles Bronson's face would likely be unrecognizable. 

So this is why we need Log (if we don't use Raw capture) - even if we end up displaying the image as an 8bit Rec709 or RGB in 8 stops of dynamic display range as above. And images like the above, without cut-off shadows or clipping highlights, are exactly what we associate with cinema-quality images as opposed to the classical video camera look.