FS7: FS7 RAW woes & data distribution between: 12-bit RAW, 16-bit RAW, & 10-bit S-Log

paulcurtis said:
In this particular case i'm coming from the point of view of actual data in front of me, from FS700 DNGs plus also similar DNGs from DBX, BMC etc,.
DNGs are integers. Most are 12 bit (all the above are).
Each has a black level set by the camera, in the case of the FS700 RAW that is around 256 (already a chunk of space gone). Then the greyscale values for each of the pixels increase to clipping. Yes the camera outputs a small S shaped curve but for our purposes lets ignore that because it really is minor.
Click to expand...
Hi Paul,
Coming from your point of view of data in front of you, linear A/D, 12 bit integers, a black level set by the camera to 256.
Assume:
A sensor A/D up-increments starting with 256 as the first count (100h).
We put the voltage start level at 256 and the voltage finish level at 4095, and up-increment the counter with 12 bit precision giving 3840 equal linear shades between 255 and 4095. Each of these shades is a linear quantity of luminance.
11.9 would be the maximum number of stops than can reside between 256 and 4095 spaced according to power-law, giving more shades to the brighter luminance values, and only 1 shade to the first stop.
 
Tom Roper said:
Hi Paul,
Coming from your point of view of data in front of you, linear A/D, 12 bit integers, a black level set by the camera to 256.
Assume:
A sensor A/D up-increments starting with 256 as the first count (100h).
We put the voltage start level at 256 and the voltage finish level at 4095, and up-increment the counter with 12 bit precision giving 3840 equal linear shades between 255 and 4095. Each of these shades is a linear quantity of luminance.
11.9 would be the maximum number of stops than can reside between 256 and 4095 spaced according to power-law, giving more shades to the brighter luminance values, and only 1 shade to the first stop.
Click to expand...

And that's exactly what we get.

The only exception being that there is a small S curve, whether that's intentional or just the nature of this sensor i don't know.

IMHO, i suspect that the ADC on the FS700 is not 16 bit and the shadows are very noisy so Sony decided it wasn't worth the effort of anything more than 12 bit.

And in fact the RAW out of the FS700 can at times be exceptional, but i can clearly see the shadow steps when grading.

My curiosity would be with why they went for baked white balance in FS7, presumably this is noise based but IMHO i think it would have been better not too. Then of course what will the FS5 have, will it be the same as the 700?

The DBX is the same

BMC do it better, each DNG frame has a 1D shaper LUT embedded which linearises the data, meaning that the raw data in the DNG is not linear but has a curve on it (it's not a log curve though). However it does mean that BMC is able to transport a better distribution in their 12 bit DNGs.

I don't actually know why it's always 12, i'd guessed that perhaps that was a DNG limitation but it's not, Lars Borg said that 16 is perfectly acceptable.

cheers
Paul
 
paulcurtis said:
I don't actually know why it's always 12, i'd guessed that perhaps that was a DNG limitation but it's not, Lars Borg said that 16 is perfectly acceptable.
Click to expand...

12 bits is plenty really, if values are distributed well.
It also helps with filesizes. 16-bit is 1/3 more data per frame. And the non-linear 12-bit compresses better: with the lack of OLPFs on BM cameras the aliased linear 16-bit wouldn't lend itself to lossless compression well. Also, their lossy DNG compression in the URSA and URSA Mini sort of needs 12-bit to be practical.
 
I'd say that 12-bits in slog is plenty for a 14 stop camera but 12-bits in linear RAW isn't enough.

So whether 12 is enough sort of depends on the format I think.
 
Well, my point it that more than 12 bits is mostly unnecessary since a proper flattening tone curve can always be applied for higher DR which is the approach taken by everyone (Arri, BM, Cineform, etc).
(except the f65?)
 
Yes I agree.

There were some publications indicating that 12-bit log is as good as 16-bit RAW. On paper I agree with that, in the real world there are trade-offs people who write articles like to ignore.
 
So, is the resolution of this debate that 12-bit Raw on a 14-stop sensor can be quite capable but Sony somehow managed to mock it up?

And that there are no third party solutions at this point?
 
Well I highly doubt Sony overlooked this to be honest.

What have other FS7 + Odyssey combo users concluded? Their input would be quite valuable right now.

When all else fails, let's see what the folks at Abelcine have to say about it from their blog:

http://blog.abelcine.com/2015/01/23/at-the-bench-recording-raw-with-the-sony-xdca-fs7/

and other sources:

In this video I don't see any black artifacts at all, but do see wild looking clippy yellow highlights:
http://www.provideocoalition.com/sony-fs7-review

https://www.youtube.com/watch?v=GcjUswh8Y9s
 
I think I've given my opinion too much on this now :)

Ultimately the camera is so good at what it does that it never need to have this crap tacked onto it.

Anyone want to buy an XDCA and 7Q for a good price? Best. Salesman. Ever.
 
I'm on my second XDCA, I sent the first one back thinking it was faulty due to the shitty internal ProRes picture quality. Turned out it wasn't faulty, and is in fact just a piece of garbage. Quality of the recording using the 7Q+ is great at normal fps, but as soon as you switch to 2K HFR it simply highlights how poor the RAW output is from the FS7.
 
Ha ha, no. As soon as I finish the doco I've bought it for, it's heading to eBay. I like the O7Q+ though. Worth the price just for the monitoring tools. Shame it doesn't get a decent feed from the FS7.
 
10-bit log RAW would be enough for 20 stops of DR (>50 values per stop of light). 12-bit linear RAW is not enough for 8 stops of DR (just 16 values in the darkest stop).
I can't believe Sony went for 12-bit linear RAW on the FS7, it would be one of those times when you add a feature but cripple it so much it actually can't do what it's supposed to do, just so you can say it's in the spec sheet. Like slog3 on the a7S II. Oh, wait...

(I'm a Sony lover right now, the last 3 cameras I've bought were Sony, but this just makes no sense)
 
Hey Samuel, you just got me thinking… a lot of confusion is also coming from the fact that the FS700 was 12-bit RAW and seemed to work ok so many consumers will be lead to believe that it will also work on the newer cameras but:

1.) I think FS700 was only 11-12 stops anyways, hence 12-bit RAW ok. (sort of)

2.) The newer Sony cams are 14 stops so if Sony went for the same internal architecture as before for handling their 12-bit RAW format, then this is epic fail.
 
Samuel H said:
10-bit log RAW would be enough for 20 stops of DR (>50 values per stop of light). 12-bit linear RAW is not enough for 8 stops of DR (just 16 values in the darkest stop).
I can't believe Sony went for 12-bit linear RAW on the FS7, it would be one of those times when you add a feature but cripple it so much it actually can't do what it's supposed to do, just so you can say it's in the spec sheet. Like slog3 on the a7S II. Oh, wait...

(I'm a Sony lover right now, the last 3 cameras I've bought were Sony, but this just makes no sense)
Click to expand...

I think we need to step back and look at the whole ecosystem.

The FS700/FS7 is less than an F5/F55 yet on paper it seems almost as capable.

I suspect that the ADC inside the camera and the supporting hardware is not running at anything like the same levels in the F5 etc,. The decision to use 12 bit RAW is perhaps because the output is so noisy down there than having 16 steps representing -5 below mid is pointless. It would just be pure noise.

Also in a linear encoding scheme the last stop will always be a single bit, it's more a question of how far below you want that last stop.

But this is true of RAW stills as well. The last stop is always 1 step. There's no magic, that's just how it is.

cheers
Paul
 
The output from the sensor has noise. You just can't put 12 stops of DR into 12-bit linear RAW, even at 1e- (which I believe is where Sony's state-of-the-art is), noise is going to eat, completely, at least a stop. And then there's the gradation issue.

IIRC, from the tests when the a7S came out, it has slightly better DR than the FS700, and I measured the a7S at 11.7 stops, so maybe 12-bit linear RAW was just-barely-enough for the FS700. It definitely should not be enough for the FS5.
 
Liam Hall said:
Ha ha, no. As soon as I finish the doco I've bought it for, it's heading to eBay. I like the O7Q+ though. Worth the price just for the monitoring tools. Shame it doesn't get a decent feed from the FS7.
Click to expand...

Yeah, I've also been pretty disappointed with my XDCA to Odyssey solution. I've used it on a couple commercials for standard FPS to ProRes, and that turned out well, but *most* of my reason for owning the two was for high frame rate, which like you said has left a bit to be desired. I haven't been comfortable enough to shoot 240fps to ProRes for a client, so I generally fall back to internal 180fps, which they love and it grades great. 99% of my work is web video where the clients aren't going to notice the difference between XAVC-I and ProRes once it's uploaded. The XDCA and Odyssey just make the FS7 into more of a hulking beast than the FS700 was!

Along with my upgrade to the A7Sii (from the original) I don't really need my 7Q+'s 4K HDMI input either. I really loved the 7Q on my FS700, it was a total game changer, and I love the screen/exposure/focus tools, but it's getting harder to justify it in my kit when I haven't been using it much lately...
 
You can fit 14 stops into 12 bit linear. It's actually quite simple, you use floating point math and that's how it's done. I believe Sony also use floating point math for the 16 bit raw of the F5/F55 as this frees up more data for the darkest stops and gives smoother shadows than would be possible with direct 1:1 16 bit encoding. The problem with floating point math though is that you end up with some degree of data rounding where a value is rounded up or down to the nearest available code value. This makes little to no difference once you get beyond about 100 code values per stop as even if you halve the number of code values the rounding is still smaller than the signal noise, so has no impact on the final image quality. But in the darker stops it can have an impact. With 12 bit linear the large amount of data rounding means that noise becomes much coarser, especially in the shadows as the noise gets rounded up or rounded down, in effect making the noise signal larger than it should really be. In addition with 12 bit there is much more data rounding in the darker stops as the steps between code values are larger and this can result in banding in the shadows.
 
Do we know for certain Sony is using floating point math for their linear RAW, and it would seem even if so it is not really a 'good' working solution?

If in the end you have to overexpose by 1 - 2 stops you're effectively using a 12 stop camera. Which is still plenty of course.
 
You must log in or register to reply here.
Back
Top