Experiments in SLog

harddrive said:
Slight correction, but I think RAW from the FS-700 will be more like that from the F5 than the F55? (I think that was Alisters original comparison, wasn't it?) I think it's quite likely the FS700 and F5 may share the same sensor, but the F55 is different - extra hardware to give the global shutter capability, and colorimetry more suited to cinema than TV, for starters.
Click to expand...
No! Alister Chapman mean exactly FS700 and F55.. not F5... ( http://www.alisterchapman.com/2013/06/25/stunning-fs700-raw-performance-really-impressive/ ) and you don't have to explain me about the difference between Rolling Shutter and Global Shutter.. and about 709 CFA and Cinema CFA :)
 
I'm always trying to find my mistakes so i did some more tests yesterday, it will take a day or so to sort them out as i'm busy today. I did some more with the step wedge but i also used a macbeth chart and did some reflectance range (like Art), although i had to use a combination of stopping down and ND to get a full range. Bit more complicated but lets see.

Also in SLog if you just record a lenscap then where are your blacks? They're elevated, meaning that you're not even using the full 8 bit range. I've not found a way to get them down as Black Level has no effect. Black Gamma does but it appears to do odd things to the shadows (I recorded those as well to see if i can understand)

I'm really not fussed about the range either. 11.5 is more fine for me.

Also Serge quite rightly points out that what's important is the RAW side and remember that Sonys RAW is linear. This is what i mean about 12bits > 12stops of range. The RAW is only ever going to give you 12 stops because it's 'linear' RAW and as i understand it - it's the linear light from the sensor. There is no Slog, no curves. Slog is a small step on the way to RAW.

If Sony is putting out 'only' 12 bit linear RAW (compared to 14 bit on the F5) then doesn't that kind of imply that's it? To suggest that you can record more dynamic range via 8bit SLog than RAW doesn't seem right to my mind.

So either the RAW isn't Linear or Sonys definition of linear is not the same as mine or the sensor really isn't doing 14 steps (or Sony has hobbled the RAW output).

Now what looks to be ideal is that the O7 at some point will take the 12 bit linear and record it as 10 bit log, so a better file size with some log encoding (of their choice, doesn't have to be sLog)

cheers
Paul
 
paulcurtis said:
. The RAW is only ever going to give you 12 stops because it's 'linear' RAW and as i understand it - it's the linear light from the sensor.
Click to expand...

I could be wrong, I'm not an engineer, but I think that assumes a linear electrical response from the chip. If the relationship was strictly linear then a 12bit A/D converter would not be able to record 14 stops of range without some form of signal processing prior to digital conversion.

However, I suspect the sensitivity of the chip becomes non-linear at both ends of its response range. i.e. as it approaches white clip, each doubling of light (e.g. each stop) does not lead to a doubling of voltage output - there is an increase, but not a doubling. Because of this, you can get more than 12 stops of exposure range within a 12bit range, as the last few stops are 'naturally' compressed into a smaller output range in the analogue realm. So long as you know the response of the chip you can interpret that information correctly to unpack the extra stops.
 
Who says the the FS700 has a 12bit ADC? And who says the 12bit RAW-over-BNC port is linear? What was interesting was that when I was playing around with the 7Q on a FS700 at Cine Gear Expo the other day, even in RAW mode, changing the picture profile appeared to change the RAW data being sent to the 7Q... I was told that the camera should be put in Slog2 mode while outputing 12bit RAW over BNC, so maybe some gamma transform is applied to the RAW data before it is sent over BNC.
 
Colinelves said:
However, I suspect the sensitivity of the chip becomes non-linear at both ends of its response range. i.e. as it approaches white clip, each doubling of light (e.g. each stop) does not lead to a doubling of voltage output - there is an increase, but not a doubling. Because of this, you can get more than 12 stops of exposure range within a 12bit range, as the last few stops are 'naturally' compressed into a smaller output range in the analogue realm. So long as you know the response of the chip you can interpret that information correctly to unpack the extra stops.
Click to expand...

Well there's a good thought - i've always understood sensors to be truly linear, unlike a film response, but perhaps there's something there. I wonder how we find out.

nyvz said:
Who says the the FS700 has a 12bit ADC? And who says the 12bit RAW-over-BNC port is linear? What was interesting was that when I was playing around with the 7Q on a FS700 at Cine Gear Expo the other day, even in RAW mode, changing the picture profile appeared to change the RAW data being sent to the 7Q... I was told that the camera should be put in Slog2 mode while outputing 12bit RAW over BNC, so maybe some gamma transform is applied to the RAW data before it is sent over BNC.
Click to expand...

The Sony RAW is Linear: http://pro.sony.com/bbsc/ssr/product-NEXFS700UK/

I don't know what the ADC is, i would have assumed 14 bit but it could be 12. It's a prosumer camera after all.. Why have RAW at 12 bit on the FS700 and 14 on the F5?

The change to SLog is for monitoring purposes.

Obviously without being able to experiment i have no idea whether any of the settings affect the RAW but i would have thought that very unlikely, it wouldn't be RAW then. Also the Sony RAW viewer does not have controls to correct for incoming gamma corrected RAW images. I would be 99.9% sure that RAW is RAW in this case, and the SLog just for monitoring.

There's a nice conversation on CML at the moment about Slog on the F55 and Art and a few others have quite rightly pointed out that Slog should be exposed correctly(ish) because the point of it is the film like highlight handling. But for us FS700 users with 8 bit RAW we're not using it for film like handling, we're just bending it so get a different curve and better highlights for us.

cheers
Paul
 
paulcurtis said:
Well there's a good thought - i've always understood sensors to be truly linear, unlike a film response, but perhaps there's something there. I wonder how we find out.
Paul
Click to expand...
It's simple physics. Light sensors are linear over their designed range of sensitivity, then diverge beyond their designed point. I'm fairly sure Sony would not publicly disclose their sensor performance curves, unless for an industry tech journal, somewhere.

paulcurtis said:
The Sony RAW is Linear: http://pro.sony.com/bbsc/ssr/product-NEXFS700UK/

I don't know what the ADC is, i would have assumed 14 bit but it could be 12. It's a prosumer camera after all.. Why have RAW at 12 bit on the FS700 and 14 on the F5?


Paul
Click to expand...

Because a 12 bit processor would be less expensive (aka cheaper) than a 14 bit processor.
 
KINOKS said:
Well raw means raw, so it's the same bit depth as the A/D, isn't it?
Click to expand...

RAW means whatever the person who said it means. It's definitely not a given that the recorded bit depth is the same as the ADC bit depth or that ADC is linear these days. My understanding is that linear ADC is by far the most common but it does seem from specs that it is a bit more complicated than bits=stops as it would be with linear ADC. Lately a lot of companies don't even publish their ADC specs. Look at the Canon C500 that outputs 10bit RAW... Is it from a linear ADC? Is it a log ADC to match the log RAW? No one seems to know, but if it gets the job done it doesn't really matter.

RAW generally just means data formatted similarly to the photosite array, which in most cases means bayer data. That data gets processed no matter what for black shading and other calibrations, and it can even be transformed/encoded/compressed to another bit depth or whatever storage format. It makes a lot of sense to transform the RAW data before storage. Sony calling it Linear RAW seems to imply it is not transformed from linear off the sensor besides basic calibrations, but who knows what it means... We can speculate all we want but probably Sony and Convergent design are the only one's who actually know what is coming out of the BNC connector in RAW mode. There is nothing that says RAW means a pixel's data must exactly match the voltage from the photosite, some integrity is of data is naturally lost in compressed RAW formats but that doesn't make it not RAW if it's still stored in a bayer sequence and each pixel stored corresponds to a photosite on the sensor array.

I've counted 14 stops with my own eyes on my own camera, so maybe its a 12bit ADC and maybe it is true 12bit linear RAW, but if it is it seems to get past the assumed limitations of a linear ADC which would have a lot of trouble differentiating the 12th stop when the FS700 does not... I don't have an explanation, and from the sound of it no one else does either, but it does seem to work and ADC bit depth hardly seems to be a DR limiter in the FS700. I'm sure seems unlikely RAW from the FS700 will be inferior in the lower stops than the 8bit AVCHD.
 
Last edited:
Video and digital photographs are not linear encoded ever. They are gamma encoded to match human visual response. Each discrete equal value gray scale step from black clip to white clip is represented by a discrete 8,10, 12 bit or higher word.

Gamma encoded images store tones more efficiently. Since gamma encoding redistributes tonal levels closer to how our eyes perceive them, fewer bits are needed to describe a given tonal range. Otherwise, an excess of bits would be devoted to describe the brighter tones (where the camera is relatively more sensitive), and a shortage of bits would be left to describe the darker tones (where the camera is relatively less sensitive):

[TD="align: left"] Original: [/TD]
[TD="align: left"]
gamma_gradient3b.jpg
[/TD]

[TD="align: left"][/TD]
[TD="align: right"] ↓ Encoded using only 32 levels (5 bits)
[/TD]

[TD="align: left"] Linear
Encoding: [/TD]
[TD="align: left"]
gamma_gradient2b.png
[/TD]

[TD="align: left"] Gamma
Encoding: [/TD]
[TD="align: left"]
gamma_gradient1b.png
[/TD]
[FONT=verdana !important]Note: Above gamma encoded gradient shown using a standard value of 1/2.2
See the tutorial on bit depth for a background on the relationship between levels and bits.
[/FONT]​
Notice how the linear encoding uses insufficient levels to describe the dark tones — even though this leads to an excess of levels to describe the bright tones. On the other hand, the gamma encoded gradient distributes the tones roughly evenly across the entire range ("perceptually uniform"). This also ensures that subsequent image editing, color and histograms are all based on natural, perceptually uniform tones.
Click to expand...

Low bit rate high compression long GOP codecs can throw out enough of the original image data that the shadows block up to the point they may look as if they were linear encoded, but that is not the same thing. They just blend fine value changes in the uncompressed data into mush.

Here is a link to an uncompressed 720p 4:2:2 8-bit broadcast AVI format clip. You may have to import it into your NLE to view it. Most video players don't support it. 42 seconds, 299MB.

https://www.dropbox.com/s/nnk0n9jzyi8slc7/edit #19.avi

Here is the 9MB Vimeo version. H.264 encoded from the above. Both are from a DNG raw work flow test I did.


Try some grading on each of the above and you will see why I loath low bit rate AVC codecs for camera originals.
 
Last edited:
From Sony papers:
The upgrade adds this to FS-700:

1. 12bit 2K RAW & 4K RAW recording to optional HXR-IFR-5 / AXS-R5 RAW recording system.
2. 4K recording up to 60fps and 2K recording up to 240fps without buffering
3. 120fps 4K RAW recording-to-buffer for about 3.74 seconds ( 17 seconds playback time in 24p )
4. Two new Gamma curves: "S-Log-2" and "800%ITU709"
5. Record trigger command embedded in the HDMI and SDI outputs..
6. Upgraded FS700 may interface with CD Odyssey 7Q recorder/monitor, which utilizes Cinema DNG RAW recording up to 240fps in 2K mode and 4K recording Odyssey 7Q creates DPX files.

When recording RAW with AXS R5 recommended: White Balance 3200K, 4300K or 5500K, and GAIN -0 ( Minimum ISO sensitivity ),
Flicker adjustment is set to 0, images recorded in 17:9 Aspect Ratio, even in viewfinder / external monitor we see only 16:9 AR (Edge crop)
Changing the Gamma Curve will change the dynamic range of the image recorded.
Picture Profile settings cannot be applied to RAW images
When "800%ITU709" or "S-Log-2" is set, following picture profile items become ineffective:
BLACK LEVEL , COLOR MODE, COLOR LEVEL and COLOR PHASE

....when I open RAW files in Sony RAW Viewer - any additional settings are coming from INFORMATION in FILE.. but they are NOT affect the RAW File directly!
if I change INPUT SETTINGS - image looks different from base "REC 709"

so ..?? :)
 
SergeSmArt said:
From Sony papers:
The upgrade adds this to FS-700:

1. 12bit 2K RAW & 4K RAW recording to optional HXR-IFR-5 / AXS-R5 RAW recording system.
2. 4K recording up to o60fps and 2K recording up to 240fps without buffering
3. 120fps 4K RAW recording-to-buffer for about 3.74 seconds ( 17 seconds playback time in 24p )
4. Two new Gamma curves: "S-Log-2" and "800%ITU709"
5. Record trigger command embedded in the HDMI and SDI outputs..
6. Upgraded FS700 may interface with CD Odyssey 7Q recorder/monitor, which utilizes Cinema DNG RAW recording up to 240fps in 2K mode and 4K recording Odyssey 7Q creates DPX files.

When recording RAW with AXS R5 recommended: White Balance 3200K, 4300K or 5500K, and GAIN -0 ( Minimum ISO sensitivity ),
Flicker adjustment is set to 0, images recorded in 17:9 Aspect Ratio, even in viewfinder / external monitor we see only 16:9 AR (Edge crop)
Changing the Gamma Curve will change the dynamic range of the image recorded.
Picture Profile settings cannot be applied to RAW images
When "800%ITU709" or "S-Log-2" is set, following picture profile items become ineffective:
BLACK LEVEL , COLOR MODE, COLOR LEVEL and COLOR PHASE

....when I open RAW files in Sony RAW Viewer - any additional settings are coming from INFORMATION in FILE.. but they are NOT affect the RAW File directly!
if I change INPUT SETTINGS - image looks different from base "REC 709"

so ..?? :)
Click to expand...

The raw file is gamma encoded to CIE XYZ color space at a reference gamma of 1. For 12 bit raw that is 4096 equal gray scale value steps from analog black clip to analog white clip. If the AD converter is properly calibrated, digital values typically max out slightly after the analog voltage max clip point to avoid nasty artifacts.
 
SergeSmArt said:
No! Alister Chapman mean exactly FS700 and F55.. not F5... ( http://www.alisterchapman.com/2013/06/25/stunning-fs700-raw-performance-really-impressive/ ) and you don't have to explain me about the difference between Rolling Shutter and Global Shutter.. and about 709 CFA and Cinema CFA :)
Click to expand...
OK - I stand corrected on what Alister actually did! (I understood he had an F5, so it seemed that was more likely the comparison - but that link makes it clear!)

In terms of what the A-D convertors etc actually are, it's worth quoting one sentence from that link: (my highlighting)
......after all these are both cameras shooting raw with very similar sensors (the FS700 quite possibly uses the same sensor as the PMW-F5). The F55 sensor has a wider colour gamut and the raw output is 16bit while the FS700 is 12bit.
Click to expand...
 
I've played, a bit, with 800% Rec 709......empirical evidence:
a- it boosts the highs, but not quite as severely as Slog2
b-it's dynamic range, in bright scenes, overfills the histogram. meaning, for a "properly" exposed shot, both the highs and lows are out of range. I conclude that this is a kind of Slog2 designed for flat, controlled lighting, as in indoors or on-stage.
 
harddrive said:
OK - I stand corrected on what Alister actually did! (I understood he had an F5, so it seemed that was more likely the comparison - but that link makes it clear!)

In terms of what the A-D convertors etc actually are, it's worth quoting one sentence from that link: (my highlighting)
Click to expand...

Sorry, maybe I don't understand UK English.. but what I read in Alister post is this:

" On the Sony booth there was a PMW-F55 with a 50mm f2 Sony PL lens next to an FS700 with the 18-200mm Sony servo zoom and this was attached to the IFR5/AXS-R5 combination for raw recording. This gave me the opportunity to grab some almost side by side raw comparison footage from both cameras. Now this is hardly a fair contest. The F55 had one of Sony’s great PL mount lenses and the FS700 had the low cost zoom. " ... " Despite the vastly different lenses the results were quite surprisingly close...."
... and , at the end: "... So what do I see? Frankly there really is very little difference between the two.... " " In pure “bang for you buck” terms the FS700 and IFR5/R5 is a very powerful combination... "

so what we are talking about ?? - about "A-D convertors" ...and what "he DID" .. or about "what he actually says" ? :Drogar-BlackEye(DBG
 
From the details that Serge posted

>Changing the Gamma Curve will change the dynamic range of the image recorded.

Well that suggests that the Sony Linear RAW isn't, either Linear or RAW, so all bets are off :)

I've looked at the FS700 and F55 RAW side by sides and no, there isn't much in it. But it's also a well controlled scene with no range that would actually tax either camera. So it's not a useful comparison or test of range.

I keep mentioning that 11/11.5 stops is a lot of range.

Razz16, i'm not sure where you're going with the bolex examples but no distribution format would be linear encoded no, but some acquisition formats can be. Also i don't think anything is saying anything about compression here. The bolex is using a KAI sensor, they're lovely sensors and best of luck to them but i don't see how this is relevant here?

Noah, would you be able to share more details of how you saw 14 stops, an approach that we could duplicate and compare?

cheers
Paul
 
Looking at the RAW Viewer workflow with sample FS700 clips, some observations.

Firstly the clips metadata has separate fields for monitoring curve and 'capture gamma equation'. The capture is SceneLinear so far, i've not seen anything other than that.
The video codec is down as F55RAWSQ which is wrong, so there must be an update due for the software.

There is metadata for camera knee point and slope, even with the gamma being SceneLinear, so this could be the key to >12 stops out of 12 bit linear, that there's a knee being applied (which to my mind doesn't make it SceneLinear anymore but anyway)

Now the purpose of the RAW viewer is to get files out to grade with.

In the input settings you can only have Material Colour Space, which appears to always be set to S-Gamut/Linear. Now S-Gamut is for the colour sensor in the F55/F65 which means that it would be wrong for the FS700 which has the same colour filters as the F5 (REC709).

Beyond that point you're tweaking the exposure and colours as a post process. You can also add a number of LUTs ready for exporting. So typically you might get out 16 bit linear ACES based DPX files, or 10 bit SLog DPX files, depending on your post workflow.

So the points to take are:

- there's a knee value in the metadata. Whether this ever changes i don't know or whether you can change it in camera. No idea.
- There is a capture gamma but i don't know whether that can be changed either.

So does anyone around here still have an R5 set up that might be able to run off some files with different camera settings to see if these are changeable?

cheers
Paul
 
I think people should just look again at Art Adams original post about the FS700 dynamic range and the gamma curves:

http://provideocoalition.com/aadams/story/cameras_sony_fs700_dynamic_range_presentation/

He's really not saying anything different from Paul. For every single cinegamma he says there are 11 useable stops, but some, e.g. Gamma 4 gives you extra 'gravy' stops (3 in the case of cine 4, making 14) - these extra stops are well into the noise floor: you cannot see them on the waveform monitor as they are lost among the noise.

But you can look at the picture of the chart and count them, they're there. You can argue as much as you like about what '12bit linear Raw' means regarding the number of stops the camera can (should) have, or you can look at the damn picture and count them! Clearly the camera can reproduce 14 stops of latitude in the right circumstances - because they are there in Art's chart.

Is it a 'useable' 14 stops? well that depends on your definition - you won't be able to discern any detail in those stops as detail will be lost in the noise, but you will get a sense of changes in tone over larger areas - hence they're 'gravy', don't put any important detail in there as it will be lost.

Those stops might also be there in Paul's test, I don't know as the image on the blog is low res and I don't know if Paul has taken the time to look at just the image to make an assessment as to whether or not he can see separate stops across his whole (13.5 stop) chart. To be honest, I doubt it as I don't think that was the intention of his test, because he's not counting the gravy stops. Paul is only counting the stops he can discern on the waveform monitor - those that are distinguishable from noise and therefore are of use.

But, at the end of the day, you can argue all you want about what the camera can and can't do, about what is and isn't possible with linear 12bit Raw, and even what 12bit linear Raw means, or should mean. Or you can just count the damn stops and be done with it.

Do I think the camera has a DR of 14 stops? Yes. Do I think those last 3 shadow stops are useable? No. They're too noisy (and it's one reason I don't like Cine 4 - as it makes to great a use of them to give you the extra highlight head room). Do I think Raw will magically give you more than 14 stops max, 11 stops useable? No. The noise is a physical feature of the sensor, you can't just magic it away.

You may be able to eke out a couple of additional clean stops by using the extra resolution in a 4K master and some temporal noise reduction to produce a cleaner HD image, but you'll be losing some of the detail in the shadow areas anyway I suspect.

But maybe someone with a Raw recorder could take the time to shoot a DR test chart like Art did and then we would know - and we could stop wasting out lives arguing about it here!
 
You must log in or register to reply here.
Back
Top