SLOG2 picture profile

[paulcurtis]

I should have made that all clearer, but Alan is right as usual.

That's why the O7 is such an important development for us, even if you work in HD only - at least compared to the expense and ergonomic hilarity of the sony solution.

cheers
Paul

 

[Razz16mm]

Well the bad news is that the FS700 only outputs 8 bit to external decks. I'm well versed what the benefits of log encoding are but on the stock FS700 you're not going to get more than 8 bit out of those ports, even if it's put into a 10 bit wrapper.

The only higher colour depth source is via RAW recording.

Within the 8 bit scale the top 100 values handle the 90% highlight and up. And as you say the bottom of the curve is super compressed. As i mentioned before there are some places where shooting and keeping the highlight values in detail can be useful, but as an everyday all over option it isn't going to work, unlike, say SLog on the F3.

Also AFAIK SLog isn't about film emulation at all, it's about mathematical compression of the linear light levels into a container that spreads them in steps that our vision system uses the best.

cheers
Paul

8-bit is not nearly as much of a problem as 4:2:0 compressed AVC. and the value encoding range you show for 8-bit is not accurate. The encoded video is linear. Equal number of bit values per stop from 0-100 IRE. 16 is black, 128 is middle gray, and 235 is white, with equal distribution of steps throughout the range. Roughly 26 grayscale steps per stop each of which is represented by an 8 bit word. This whole concept that one stop change is equal to half the total remaining bit value range is wrong. That is not how video is encoded. 4:2:2 8-bit uncompressed from the HDMI or SDI output has a lot more data to work with than AVC. A 4:2:2 8-bit uncompressed broadcast AVI file for a 42 second clip I was playing with is over 300MB of data. The AVC file is 12MB of data. AVC compression cost you 95% of your original 8-bit encoding.

 

[paulcurtis]

8-bit is not nearly as much of a problem as 4:2:0 compressed AVC. and the value encoding range you show for 8-bit is not accurate. The encoded video is linear. Equal number of bit values per stop from 0-100 IRE. 16 is black, 128 is middle gray, and 235 is white, with equal distribution of steps throughout the range. Roughly 26 grayscale steps per stop each of which is represented by an 8 bit word. This whole concept that one stop change is equal to half the total remaining bit value range is wrong. That is not how video is encoded. 4:2:2 8-bit uncompressed from the HDMI or SDI output has a lot more data to work with than AVC. A 4:2:2 8-bit uncompressed broadcast AVI file for a 42 second clip I was playing with is over 300MB of data. The AVC file is 12MB of data. AVC compression cost you 95% of your original 8-bit encoding.

Sorry, i don't agree with you there.

The encoded video is obviously not linear, it's slog. The source real world is linear, the encoding is log and after an appropriate LUT is applied the video is REC709. I've not found a LUT from SLog direct to Linear, only to various 709 (which in turn can be LUT'd to Linear). So you can unfold the slog to linear but the distribution of luma values is what we're discussing.

8 bit is a problem as can be seen if you work with the footage. The shadows and mids are very low in the recorded range compared to the other profiles and suffer quite a lot of noise. It is generally considered that 10bit is the minimum for working with any log footage. The only other 8 bit log variant is Canons and that's not considered 'proper' log because it's tweaked to work in 8 bit.

You can see from this pdf from Sony themselves http://pro.sony.com/bbsccms/assets/files/mkt/cinema/solutions/slog_manual.pdf

That middle grey (18%) is around 96, and 90% is 160 (pretty much where i said it was originally). 10% of your highlight range is spread in 40% of your available bucket.

I've also done some tests myself which i'm in the process of writing up which also shows the distribution of light values visually. The 700 is actually SLog2 and i've not found any specific sony docs to see what the differences in the curves are but i'm assuming they're pretty close.

The 4:2:0 vs 4:2:2 has been done a lot on these forums and yes, there's a difference but it's more to do with grading and motion rather than resolution.

Perhaps i'm not understanding what you're saying or perhaps i'm misunderstanding something?

cheers
Paul

 

[cuervo]

hmmmm....
It was always my impression that the way human eyes experience light is in a log fashion...not linear at all. And the way film experiences light is in a log fashion. But not the same log response as human eyes.....film has a head and a tail in the log curve. This head and tail was a technology limitation of film that proved to be beneficial in blending the way that film saw light. The problem with light sensors is that they are primarily linear detectors, over the range of visible light, anyway, with no natural head or tail rolloff. Light sensors have the additional characteristic of changing their light response curve depending on the voltage applied across them. So gamma curves were invented to convert the linear way of sensor seeing to the log way of human eyes. Now, along comes Slog and Slog2, which really isn't invented to help our eyes see a scene in a natural way. The log gamma or Slog is devised to cram more useful bits into the highlights, as Paul said, not to more accurately replicate human vision. So, if one needs to replicate human vision in a scene, we need to "de-Slog" it to get it back to human log.

And all this LUT stuff is simply a contrived way to adjust the data captured by a linear camera detector to the way our eyes perceived it, in the first place. And Slog is a contrived way to capture more data in the useful region, not really designed to replicate the log of human eyes.

 

[Razz16mm]

hmmmm....
It was always my impression that the way human eyes experience light is in a log fashion...not linear at all. And the way film experiences light is in a log fashion. But not the same log response as human eyes.....film has a head and a tail in the log curve. This head and tail was a technology limitation of film that proved to be beneficial in blending the way that film saw light. The problem with light sensors is that they are primarily linear detectors, over the range of visible light, anyway, with no natural head or tail rolloff. Light sensors have the additional characteristic of changing their light response curve depending on the voltage applied across them. So gamma curves were invented to convert the linear way of sensor seeing to the log way of human eyes. Now, along comes Slog and Slog2, which really isn't invented to help our eyes see a scene in a natural way. The log gamma or Slog is devised to cram more useful bits into the highlights, as Paul said, not to more accurately replicate human vision. So, if one needs to replicate human vision in a scene, we need to "de-Slog" it to get it back to human log.

And all this LUT stuff is simply a contrived way to adjust the data captured by a linear camera detector to the way our eyes perceived it, in the first place. And Slog is a contrived way to capture more data in the useful region, not really designed to replicate the log of human eyes.

Yes, human vision responds to a power density transfer function expressed as a ratio of two values. The formula is 20log(2/1) for a doubling of measured light in foot candles or 6dB per photometric stop (f/stop). The response of a sensor is a voltage rather than power response or 10log(2/1) function. Inverse gamma correction of V/.45 is applied to sensor analog voltage output to match the power response curve of human vision. But digital encoding is just a numeric scale applied to a range of analog values and can be defined any way you want it to, linearly or logarithmically. Audio is digitized in log scale, video is digitized in linear scale to preserve good value discrimination through the full range. There are equal numbers of discrete gray scale steps encoded per analog photometric stop from black clip to white clip. Open a standard log grayscale test pattern and read the digital values of each step. You will see what I mean.
Most of the degradation of visual detail in low value ranges with long GOP cameras are the result of AVC compression, not 8-bit encoding. Uncompressed 8-bit video looks very different.

The REC709 standard defines an analog linear photometric range of 10 stops between black clip and white clip, period. Doesn't matter what bit depth the video is encoded to. A 1 meter ruler can be read in meters, decimeters, centimeters, or millimeters, but whichever you choose the length of the stick is the same. Luts are just a way of compressing broader dynamic range that cameras can capture to fit within video limits. There are literally infinite variations on how to do that. It is like a photographer dodging and burning film prints in a darkroom to preserve the important values in his image. To make matters even more confusing, none of our display technologies can actually reproduce a full linear 10 stop range, the gamma compression (2.4) applied in REC709 effectively reduces the 10 stop range to around a 4-5 stop range of actual light output as measured off the display with a light meter. So the analogy between a film negative and a paper print is even closer than one might think.

http://www.rags-int-inc.com/PhotoTechStuff/TonesnZones/

 

[cuervo]

Kinda like an Einstienian version of the Ansel Adams Zone VIII exposure system with Log-Log and semi-Log thrown in the pot and stirred briskly!:2vrolijk_08:.

 

[Razz16mm]

Kinda like an Einstienian version of the Ansel Adams Zone VIII exposure system with Log-Log and semi-Log thrown in the pot and stirred briskly!:2vrolijk_08:.

Yup. And the 0-100 IRE waveform scale is the electronic version of an Adams zone chart. The technologies may change but the underlying physics don't.

 

[paulcurtis]

The REC709 standard defines an analog linear photometric range of 10 stops between black clip and white clip, period. Doesn't matter what bit depth the video is encoded to. A 1 meter ruler can be read in meters, decimeters, centimeters, or millimeters, but whichever you choose the length of the stick is the same. Luts are just a way of compressing broader dynamic range that cameras can capture to fit within video limits. There are literally infinite variations on how to do that. It is like a photographer dodging and burning film prints in a darkroom to preserve the important values in his image. To make matters even more confusing, none of our display technologies can actually reproduce a full linear 10 stop range, the gamma compression (2.4) applied in REC709 effectively reduces the 10 stop range to around a 4-5 stop range of actual light output as measured off the display with a light meter. So the analogy between a film negative and a paper print is even closer than one might think.

http://www.rags-int-inc.com/PhotoTechStuff/TonesnZones/

Hi Razz,

Would you be able to explain where you got those values from (10 stops)? I've seen numerous stats quoted for REC709, from 5 stops upwards, with 6 being most commonly quoted. I've sat through seminars with Charles Poynton on other colour space issues but very rarely see hard facts about REC709 it's one of those standards that people quote without any reference and i'm keen to understand more.

thanks
Paul

 

[KINOKS]

...none of our display technologies can actually reproduce a full linear 10 stop range, the gamma compression (2.4) applied in REC709 effectively reduces the 10 stop range to around a 4-5 stop range of actual light output as measured off the display with a light meter.

But this differs from display to display right? This is expressed in the contrast ratio or something like that?

 

[paulcurtis]

I believe i read that the new Dolby HDR monitors can go from 40 to 4000 nits in range. That is painfully bright. I'm not sure we'd want our displays showing that degree of range. Instead we want them to compress the real world to a more comfortable viewing range. Imagine in the cinema watching a sunset and the brightness would mean we couldn't actually look at the image!

We do want our displays to be able to reproduce fine enough detail so we can appreciate the presence of that range.

cheers
paul

 

[nyvz]

You can see from this pdf from Sony themselves http://pro.sony.com/bbsccms/assets/files/mkt/cinema/solutions/slog_manual.pdf

That middle grey (18%) is around 96, and 90% is 160 (pretty much where i said it was originally). 10% of your highlight range is spread in 40% of your available bucket.

Where are you getting that? That document doesn't seem to say that.

Also remember "linear" means different things. It seems it can be used to mean linear light (luma=lux) or linear video (gamma encoded, not-log-encoded) which are very different things.

 

[paulcurtis]

Where are you getting that? That document doesn't seem to say that.

Also remember "linear" means different things. It seems it can be used to mean linear light (luma=lux) or linear video (gamma encoded, not-log-encoded) which are very different things.

In Fig 1 'Characteristic of S-Log' it shows middle grey at a code value of 394 and 89.9% white at 636 in 10 bit scales. I've seen it at similar places on other graphs too (I think abelcine had some nice ones)

I don't think Linear Video is an accepted term these days is it? Linear should always mean luma=lux, at least in vfx and post worlds...

cheers
Paul

 

[Gene Sung]

I can't contribute any real technical knowledge to this thread because most of you are far more technical than I am. But here's a test I filmed of some kids late in the day using S-Log 2. Just used the internal codec. Like everyone else, waiting for the 7Q. Shot rig-less handheld + barebones, had to work fast with mostly zooms because I only had the two sets of kids for about 1 hour.

Mild grade, exposed in the middle of the histogram, Zebras at 100% to avoid under-exposing. Under exposing is punishing. The 2nd shot of the family was under exposed and it was noise and macro block city . Now that I know how to expose for it, I LOVE S-Log 2. Not just the way better highlights, but an X-Factor as well. It just feels better IMO, as it has this luminous quality that just was not there before.

 

[Razz16mm]

In Fig 1 'Characteristic of S-Log' it shows middle grey at a code value of 394 and 89.9% white at 636 in 10 bit scales. I've seen it at similar places on other graphs too (I think abelcine had some nice ones)

I don't think Linear Video is an accepted term these days is it? Linear should always mean luma=lux, at least in vfx and post worlds...

cheers
Paul

The terms can be confusing and are often misused. Linear video refers to a linear digital encoding at a gamma of 1 without any further compression or manipulation. It is still gamma corrected for human log visual response. Linear light, which is what defines the raw sensor analog voltage response is photons/conversion efficiency = electron volts. Linear light is also related to direct value measurement in foot candles, lumens, or lux by a meter. Measuring in EV or ISO referenced f/stops is log compensated to a gamma of 1.