Experiments in SLog

[Razz16mm]

Raw will not change the native DR of the camera. It frees you from all the futzing about with the camera engineer's built in video luts. When you have full access to the "linear" ( 1 gamma) DR range unconstricted by the the 10 stop linear (1 gamma) range defined by REC709 you can modify the raw data to fit the video space any way you see fit. You choose precisely where to clip shadows and highlights, how the gray scale values are mapped through the video range, how your highlights roll off, what detail and noise level you want to achieve in the shadows. You define the gamma compression curve to suit yourself with a properly calibrated monitor as your reference. It is about creative freedom to manipulate the images.
It also can make shooting in the field much easier as you are only concerned with getting optimum exposure, everything else that defines how the image looks is done in front of the camera and in post with meta data. The processing that the camera does to output HD video with Cinegamma or S-log or whatever is bypassed, you do it on your computer in grading.

Those of us who are used to grading raw still photographs love this freedom, but it still seems to give many video shooters the willies. There is definitely a steep learning curve if you haven't done much of it.

 

[Colinelves]

Yup. I appreciate that, it's just driving me a bit mental that people are saying that the camera simply cannot have more than 12 stops of latitude because Sony are saying that Raw is only 12bits linear - ignoring the fact that a) the camera has been demonstrated to have more than 12 stops of latitude (it has, they may not all be useable, but they are there) and b) what Sony are calling Raw 12 bit linear could, in all probability mean something totally different from what they think it does.

Red say their cameras shoot 'Compressed Raw' and that, clearly, is not just 'true' Raw because it is compressed and, ergo, has undergone some post processing. Are we now saying that the Epic is not a 'Raw' camera?

Also, I think it would be pretty mental for Sony to crippled the latitude of the Raw output of the camera relative to the 8bit footage it usually shoots, so it is no great stretch of the imagination to think that they may do a little post processing to take a 14 bit A/D Raw signal and convert that to 12bit linear 'Raw' signal (maybe doing a one or two other things in the process, who knows) for output and recording. Which, by the sounds of it, is what Canon does with their 10bit 4k 'Raw' (which delivers 12 stops of latitude, as if by magic).

Which is to say: maybe, just maybe, '12bit linear Raw' is just a name - it is not strictly 'Raw' it's just more Raw than what you get already. You know, like the way they say the FS700 shoots 240fps in 'Full HD' when we all know that it's not really HD since it's based on a 2K bayer and hence, once debayered, is somewhat less than HD in resolution.

 

[Razz16mm]

Dynamic range is defined by the analog voltage output of the sensor between full well clip point maximum voltage and the noise floor voltage. Digital encoding does not define dynamic range, it just applies a numeric scale to a range of analog values. Whether you encode with 8 bits or 16 bits, the analog value range does not change. You just get finer resolution of value differences between steps.

All photographic systems are analog. You input analog light values at one end and you get analog light values out at the other end. Digital encoding is just a means of storing discrete numeric values that describe the light input and light output values. The values themselves are always analog referenced to the camera sensor on the input side and to the analog light dynamic range of a defined standard display at output.

 

[paulcurtis]

I spoke to an engineer earlier and he said he'd try to get information from japan. He's not sure how much can be revealed publicly. But lets see.

Lets assume a theoretical 14 stops (although only Arts test seems to show anything concrete here, but it is from a very reputable source), either Sony has hobbled the 700 with the RAW output to a true linear signal and 'only' 12 stops, whereas the 8bit AVC with a curve could show more range(!). Or that RAW signal isn't linear after all. The metadata for the RAW files do suggest a knee and slope, so this is entirely possible.

We just need to find someone with an R5 and an hour or two to spend doing some tests...

This is what happens when my main workstations are rendering away. Too much time pondering.

I will try denoising my footage and look at the reflectance stuff as soon as i can as well. To hunt for the mythical lost stop and a half. Maybe the higgs boson will be found first

cheers
Paul

 

[cuervo]

"This is what happens when my main workstations are rendering away. Too much time pondering."

hahahahaha...yeah, I know what you mean.

So so much of all this hand wringing and educated guessing is coming from everyone with an 8-bit camera. And with all due rspect to the people like Art or Philip or Alistair, we've yet to see user tests, out of the lab. Except for Sergesmart, but that was a simple "look at my video clip". No one with the ability to record FS700 RAW has stepped up to the plate with anything but unmeasured observation. It's going to be REALLY interesting when some real experience reports start coming in from O7Q users.

 

[dashdot]

Dynamic range is defined by the analog voltage output of the sensor between full well clip point maximum voltage and the noise floor voltage. Digital encoding does not define dynamic range, it just applies a numeric scale to a range of analog values. Whether you encode with 8 bits or 16 bits, the analog value range does not change. You just get finer resolution of value differences between steps.

All photographic systems are analog. You input analog light values at one end and you get analog light values out at the other end. Digital encoding is just a means of storing discrete numeric values that describe the light input and light output values. The values themselves are always analog referenced to the camera sensor on the input side and to the analog light dynamic range of a defined standard display at output.

I would have to agree with Razz16mm, in my limited knowledge bit depth is not related to dynamic range too much. Dynamic range is the difference between your noise floor and the electron well depth of the sensor. Some of the DSLRs allow choosing of the bit depth recorded, and you only see a change in the graduations recorded, not in the dynamic range.

I am sure there is some input from the ADC in altering the outputs, but I don't think it is the determining factor for dynamic range. I found that bythom's description of how a sensor works is quite useful in understanding what goes on. It is for stills, but a large part of it also applies for video too.

 

[paulcurtis]

I would have to agree with Razz16mm, in my limited knowledge bit depth is not related to dynamic range too much. Dynamic range is the difference between your noise floor and the electron well depth of the sensor. Some of the DSLRs allow choosing of the bit depth recorded, and you only see a change in the graduations recorded, not in the dynamic range.

I am sure there is some input from the ADC in altering the outputs, but I don't think it is the determining factor for dynamic range. I found that bythom's description of how a sensor works is quite useful in understanding what goes on. It is for stills, but a large part of it also applies for video too.

If the recording format is true Linear data then that has a direct link to the dynamic range that can be recorded through it, regardless of the sensors range.

Our musing here are

a) Can the FS700 sensor produce 14 stops and produce that in a discernible fashion
b) Is the 12 bit Linear recording format truly linear or has it a curve or knee to allow >12 stops of range

Both things that testing with a suitable setup should be able to resolve (and asking around at Sony in case there's some hard and fast limitations)

cheers
Paul

 

[paulcurtis]

A small update.

I heard back from Sony. They cannot reveal the specs of the ADC (fair enough). The SLog2 range they specify as 1300%, which is 13 stops above 18% grey. But that's the potential container rather than the actual camera.

The bad news is the knee and slope in the RAW meta data refers to the monitoring curve. But i'm still trying to clarify the 12 bit linear is just 12 bit linear part.

On another note i just found an updated Art Adams post with 4K and range

http://www.dvinfo.net/article/acquisition/sonyxdcam/fs700-the-4k-story-from-my-perspective.html/4

He's moved to this site now.

On the charts on this page (using the RAW output) he's saying that if noise is flattened there's about 11 stops visible. He's still saying 14 stops but it is all mush (i can't see the image or waveform he is using the judge for myself but the noise is very very difficult to separate). To be honest i think this pretty much concurs with what i see.

The trick to maximising the range is going to be managing the noise floor. And i would expect that 4K + Denose + Scale will give the best results.

Also my observations about the non log'ness of SLog is because this is SLog2 and not the original SLog. SLog2 has more contrast and is closer to the original scene, so that makes sense.

cheers
Paul

 

[SKYLIKE]

1300% is about 6,5 stops above 18% grey.

 

[KINOKS]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

 

[paulcurtis]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

I think you'll find that in real terms these other cameras have less 'real world' range too, as do all the other cameras out there. There is no official figure for range for any of these cameras AFAIK and that's probably because it's scene dependent as well.

Sam (Morgan Moore) pointed out something just now.

For example:

In Arts tests he choose Green as his luma because it was cleanest. I chose Red for mine because it's a more difficult colour. So immediately this would skew the results and give me more noise then him. Hence perhaps 'seeing' a bit more range.

But in the real world, the colour channels do clip independent of each other, outside the blue will clip faster and inside the red will. This will effect the real range too which makes these kinds of tests no more than a curiosity or a comparison. i.e. if you've got a bundle of cameras and you use the same test on all them you'll find relative performance but not necessarily absolute performance.

I guess that's the danger of doing tests like this - people take away absolute performance. I was investigating the idea that SLog gave more range than the hypergammas, a relative performance issue.

I personally think the 700 is an incredible camera for the money and i suspect used correctly it will match the F5 in a lot of cases, especially with the RAW output.

cheers
Paul

 

[Colinelves]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

No, because you're not comparing like with like. If you actually compare the results of both cameras against exactly the same test (e.g. The same chart, looking at the waveform output) you'll see for both the FS700 in slog2 and the F3 in slog it is step 12 that merges with the noise floor ie. both 'only' give you 11 clean stops. The difference between the F3 and the FS700 is that the 700 gives you 3 more noise stops that you can read off the image, but are lost below the noise floor in the waveform, while the F3 gives you 2 more.

F3: http://blog.abelcine.com/2011/08/04/f3-s-log-part-1-on-the-charts/

http://www.dvinfo.net/article/acquisition/sonyxdcam/fs700-the-4k-story-from-my-perspective.html/4

th pse are from the F3

 

[Colinelves]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

A small update.

I heard back from Sony. They cannot reveal the specs of the ADC (fair enough). The SLog2 range they specify as 1300%, which is 13 stops above 18% grey. But that's the potential container rather than the actual camera.

cheers
Paul

I don't think it is. The percentage refers to reflectance, with 100% being matte white. With the usual Middle grey is 18%, one above is 36%, two is 72%, (two and a half is 100%) three is 144%, four 288%, five 566%, six 1132% - so 1300% is about six and a bit stops about 18% grey.

At least that is what I'm garnering from here: http://blog.abelcine.com/2013/01/18/sonys-s-log2-and-dynamic-range-percentages/

Although they say "Sony is saying that their cameras can capture a brightness value some 13-14 times greater than that 90% white value when exposed correctly." I think this is confusing - 14 is 2 to the power 3.74 ie '14 times greater' probably should be interpreted as just under 4 stops above 90% white (which is 2 and a bit above 18% grey) - which is consistent with above.

 

[Colinelves]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

1300% is about 6,5 stops above 18% grey.

Sorry, I missed that earlier... (What he said)

 

[Colinelves]

11 stops it is then... Looks like the F3 and the F5 both surpass the FS700 in DR. This was my biggest question.

But in the real world, the colour channels do clip independent of each other, outside the blue will clip faster and inside the red will. This will effect the real range too which makes these kinds of tests no more than a curiosity or a comparison. i.e. if you've got a bundle of cameras and you use the same test on all them you'll find relative performance but not necessarily absolute performance.

Paul

I agree with Sam there. We could do with another Zacuto shootout - personally I found them to be the most informative tests.

 

[morgan_moore]

I thought the Zacuto tests were 'useless' at comparing cameras - as the DPs could add more fill light. What they proved is that given enough lights you can make a narrow DR camera look good.

In the real world we rarely have enough lights!

As for those step graphs I also think they are pretty useless - because it comes down to visual aesthetics which is in my opinion - and only I can have my opinion on my visual aesthetic - those charts dont show tonal banding or colour shifts

I post an image from a camera examining the tonality and colour accuracy of 'roll off to clipped' - to me it is ugly - to others it would be acceptable.



Anyway if that is happening in the top step of your graph (and you find it unacceptable aesthetically and stop down to avoid it) you lose another stop of usable DR, again at the bottom end 'noise' is acceptable at different levels/scenes/story contexts.. Zero dark thirty is different from a fashion spot (maybe) in terms of aesthetically acceptable noise level

My opinion is that testing is hard because one wants 'lab' conditions, but also can mis stuff in a lab (using currently accepted test methodologies) that becomes blatantly obvious once you walk out of that lab and point a camera at something real.

Or you could say that the current test methodologies are flawed if they represent the real user experience so badly.

After buying the FS100 from Abels test reports I can only say that for more costly cameras I blag/rent them and do my own tests before handing over my hard earned...

 

[KINOKS]

No, because you're not comparing like with like. If you actually compare the results of both cameras against exactly the same test (e.g. The same chart, looking at the waveform output) you'll see for both the FS700 in slog2 and the F3 in slog it is step 12 that merges with the noise floor ie. both 'only' give you 11 clean stops.

Not really. From the Abel Cine test the F3 puts out 12 clean stops while the FS700 puts out 10 clean stops (judging from Art Adams Cine 4 chart). The FS700 11th stops is further down in the noise floor then the F3s 12th stop.
I mean, this makes sense to me as F3 has bigger pixels then the FS700. I know FS700 is newer technology but it's not That much newer...



This F3 test has a LUT applied but even when comparing to a non LUTed chart (the one you posted) the result's are the same 12 usable stops + 1 semi usable. By Art Adams logic the F3 would have 13 usable stops. The 11th stop of the FS700 is really down there in the noise floor. Much like the 13th stop is on the F3...

 

[KINOKS]

Art Adams charts are not done with the Slog2 however...

 

[morgan_moore]

..but then I can take that image and using Resolve make the clip points even. and that makes testing of the cameras super hard!



Which in the scopes looks like this which I think is a far more sensitive test of a camera than one stop steps!






S

 

[paulcurtis]

Colin, you're absolutely right, i'm mixing up my 1300% with 13 stops overall. Doh. Thanks for the sanity check. Sony has always maintained 6/6.5 over mid grey for SLog2 and that's pretty much what i've seen. It's the under that becomes more difficult to tackle.

The F3 has incredible range, especially with those big pixels. However noise is also a function of resolution and averaging, i would expect to be able to pull better de-noised frames out of 4K source than a denoised 1080p frame for example.

Sanjin,

When comparing the absolute top and bottom of these charts we're heading into splitting hair territory as i tried to say. Factors such as white balance, lighting and the scene will all affect how the extremes react. For example on Arts charts for the F3 - where are those lines coming from? A Luma channel from the original YUV output? Or a reconstruction of R, G or B? Any of these sources will change the result. Hence like for like, in the same test, with the same camera etc,.

Sam is along the right lines and maybe we should make Sam the designated camera tester!

But in seriousness, what technique would people find suitable? On CML the tests are usually a macbeth with under and over, which makes sense. The step wedge and Arts chart just means faster results, but i always had an issue with reflectance charts because of their nature. I've always though a transmission test is more valid.

cheers
Paul