Topics

Red noise in Canons worse in ACES workflow?

Geoff Boyle
 

I'm testing C200 at the moment at 2500 ISO for some covert sequences in a movie, I'm getting noise in the red channel that is pretty bad.
I thought it was just the camera we had hired in for the test but going back to my original test files there looks like there's noise in the reds in all the C series cameras when pushed.

Anyone lese?

Geoff

Bill Hogan
 

HI GEOFF;

Is your lighting Tungsten or Daylight?  Are the CANON cameras using a Tungsten or Daylight balanced sensor?
If it is a Daylight sensor you will certainly have more red noise.

Thanks Bill Hogan
Burbank, CA

On Fri, Mar 9, 2018 at 1:28 AM, Geoff Boyle <geoff@...> wrote:
I'm testing C200 at the moment at 2500 ISO for some covert sequences in a movie, I'm getting noise in the red channel that is pretty bad.
I thought it was just the camera we had hired in for the test but going back to my original test files there looks like there's noise in the reds in all the C series cameras when pushed.

Anyone lese?

Geoff



Geoff Boyle
 

It’s a daylight sensor which is why I’m lighting it with daylight 😊

 

Cheers

 

Geoff Boyle NSC FBKS

Cinematographer

Zoetermeer

www.gboyle.co.uk

+31 (0) 637 155 076

 

From: cml-post-vfx-aces@... <cml-post-vfx-aces@...> On Behalf Of Bill Hogan
Sent: 09 March 2018 10:50
To: cml-post-vfx-aces@...
Subject: Re: [post-vfx-aces] Red noise in Canons worse in ACES workflow?

 

HI GEOFF;

 

Is your lighting Tungsten or Daylight?  Are the CANON cameras using a Tungsten or Daylight balanced sensor?

If it is a Daylight sensor you will certainly have more red noise.

 

Thanks Bill Hogan

Burbank, CA

 

On Fri, Mar 9, 2018 at 1:28 AM, Geoff Boyle <geoff@...> wrote:

I'm testing C200 at the moment at 2500 ISO for some covert sequences in a movie, I'm getting noise in the red channel that is pretty bad.
I thought it was just the camera we had hired in for the test but going back to my original test files there looks like there's noise in the reds in all the C series cameras when pushed.

Anyone lese?

Geoff

 

 

 

 

Art Adams
 

I'm curious how we know it is a daylight-balanced sensor.

--
Art Adams
Director of Photography
San Francisco Bay Area

David F. - Ladigitaltech
 

There was a cinematographer who got yelled out by a famous actor Cristian Bale, he did a test on the Canon DSLR's and mentioned whole number iso bump ups resulted in a lot of noise vs 1/3 bump ups such as 1000 2000 2500 had more noise than 1250-3200 maybe you can test that out on the c200?

Thanks Dave

On 3/9/18 6:32 AM, Geoff Boyle wrote:

It’s a daylight sensor which is why I’m lighting it with daylight 😊

 

Cheers

 

Geoff Boyle NSC FBKS

Cinematographer

Zoetermeer

www.gboyle.co.uk

+31 (0) 637 155 076

 

From: cml-post-vfx-aces@... <cml-post-vfx-aces@...> On Behalf Of Bill Hogan
Sent: 09 March 2018 10:50
To: cml-post-vfx-aces@...
Subject: Re: [post-vfx-aces] Red noise in Canons worse in ACES workflow?

 

HI GEOFF;

 

Is your lighting Tungsten or Daylight?  Are the CANON cameras using a Tungsten or Daylight balanced sensor?

If it is a Daylight sensor you will certainly have more red noise.

 

Thanks Bill Hogan

Burbank, CA

 

On Fri, Mar 9, 2018 at 1:28 AM, Geoff Boyle <geoff@...> wrote:

I'm testing C200 at the moment at 2500 ISO for some covert sequences in a movie, I'm getting noise in the red channel that is pretty bad.
I thought it was just the camera we had hired in for the test but going back to my original test files there looks like there's noise in the reds in all the C series cameras when pushed.

Anyone lese?

Geoff

 

 

 

 


-- 
David - info@... http://ladigitaltech.com 310-540-7859

Justin Johnson
 

You could be seeing out-of-gamut colors with negative values creating artifacts when viewed through your output transform. 

ACES primaries are positioned to encompass the human spectral locus.  Not the spectral locus of CMOS sensors.  Canon Cinema Gamut is positioned in a way to cover the spectral response of their sensors.   Noise can easily step way outside of gamut in ACES and my guess is that you are seeing some artifacts related the resulting negative values.  You could try shooting the same test with Cinema Gamut. 





On Fri, Mar 9, 2018 at 1:28 AM, Geoff Boyle <geoff@...> wrote:
I'm testing C200 at the moment at 2500 ISO for some covert sequences in a movie, I'm getting noise in the red channel that is pretty bad.
I thought it was just the camera we had hired in for the test but going back to my original test files there looks like there's noise in the reds in all the C series cameras when pushed.

Anyone lese?

Geoff

Art Adams
 

I believe that's a different thing. If I recall correctly, the full stop bumps in DSLR ISO are analog gain adjustments, while the 1/3 increments are positive and negative digital gain adjustments to each analog gain adjustment.

I've not seen that in any of the Canon cine cameras I've used. They do seem a bit noisy to me, though, 

--
Art Adams
Director of Photography
San Francisco Bay Area

Art Adams
 

>Noise can easily step way outside of gamut in ACES and my guess is that you are seeing some artifacts related the resulting negative values.

Okay, this is good stuff. :) When you say negative values, do you mean numerically negative or simply outside of gamut? And numerically in which axis?

Would a workflow outside of ACES potentially eliminate the noise in this case, as there would be a different transform to the viewing environment?

And wouldn't an ACES workflow take the Canon Cinema Gamut into account so something like this wouldn't happen? This seems like a fairly serious issue, although I guess if it only happens with extreme noise then maybe that's just exceeding the limits of that workflow.

Thanks.

--
Art Adams
Director of Photography
San Francisco Bay Area

JD Houston
 

On Mar 9, 2018, at 10:36 AM, Art Adams <art.cml.only@...> wrote:

When you say negative values, do you mean numerically negative or simply outside of gamut?
They are the same thing. If any one, or two components are negative, it means a color is positioned outside of the gamut.
If all three are negative, it is negative luminance color, and while that can happen with the math, it is not common.



Would a workflow outside of ACES potentially eliminate the noise in this case, as there would be a different transform to the viewing environment?
Desaturating the source to get rid of the extreme colors will also cause the noise to be reduced. Saturation can then be selectively reapplied (with a 6-axis secondary).

And wouldn't an ACES workflow take the Canon Cinema Gamut into account so something like this wouldn't happen? This seems like a fairly serious issue, although I guess if it only happens with extreme noise then maybe that's just exceeding the limits of that workflow.
Ideally, the IDTs for a camera being converted into ACES do not pass non-real colors. The code values for some cameras produce values that are outside the spectral locus (and thus outside of ACES.)
These values can get clipped and will show as noise. While S-Cine is on the Red side, Alexa as a Blue issue with their WCG. One way to fix is to have a better mapping in the IDT that does not allow negative numbers to get through.



Jim Houston
Consultant, Starwatcher Digital, Pasadena, CA

Justin Johnson
 


On Fri, Mar 9, 2018 at 10:36 AM, Art Adams <art.cml.only@...> wrote:
>Noise can easily step way outside of gamut in ACES and my guess is that you are seeing some artifacts related the resulting negative values.

Okay, this is good stuff. :) When you say negative values, do you mean numerically negative or simply outside of gamut? And numerically in which axis?


When colors go outside of gamut they produce negative values if they are in a floating point format.   The color primary positions define a point in Yxy space that is 100% saturated.   And if you have, for instance, a red color value that is outside of a color space's triangle, it is going beyond 100% saturation, so green and blue values must go below 0 (or clamp at 0) to compensate.  This results in data loss if those negative values get clamped, and either way, can produce strange artifacts when viewed through a particular output transform. 

 

Would a workflow outside of ACES potentially eliminate the noise in this case, as there would be a different transform to the viewing environment?

Potentially, but I would need to know more specifics about the data flow to say for sure.  Many have been using LMT's to fix artifacts by adjusting the primary values to reign in out-of-gamut values.
 

And wouldn't an ACES workflow take the Canon Cinema Gamut into account so something like this wouldn't happen? This seems like a fairly serious issue, although I guess if it only happens with extreme noise then maybe that's just exceeding the limits of that workflow.

In my expreience, you are likely to see artifacts with ACES on any super saturated values, not just noise.  My belief is that IDTs are supposed to handle this, but their creation is up to the camera vendors and my understanding is that they are usually just matrix transfroms which can push data out-of-gamut.   This is where I yield to those who know more about the subject than I do.   I think the goal of an IDT is to remap the sensor data into the human spectral locus and thus ACES, but again, my understanding of ACES isn't complete.

As you can see below, it seems camera makers are choosing to define their color spaces to encompass or to be drawn near the spectral loci of their sensors, which due to the nature of CMOS sensors is different than the human spectral locus. 

 

David F. - Ladigitaltech
 

I tested it out myself and saw the same scene and I saw more noise and it was a standard scene, so easy for anyone to test it out grab your camera and take a shot with two iso's and play back on a monitor at 100-200% and you'll see if it does or not... it's not like it will be a make or break situation but it will be somewhat noticeable... my personal opinion is to shoot a person with blonde/brown hair and look at that, I've noticed you can really see noise in hair...

Good luck...

On 3/9/18 10:29 AM, Art Adams wrote:
I believe that's a different thing. If I recall correctly, the full stop bumps in DSLR ISO are analog gain adjustments, while the 1/3 increments are positive and negative digital gain adjustments to each analog gain adjustment.

I've not seen that in any of the Canon cine cameras I've used. They do seem a bit noisy to me, though, 

--
Art Adams
Director of Photography
San Francisco Bay Area


-- 
David - info@... http://ladigitaltech.com 310-540-7859

Jack Jones Colourist
 

From a colourist perspective, when I’m using Nucoda, you have the option to clamp, process or preserve negative values using their gamma matrix tool.

You could do this by using clipping tools such as soft clips or curves in other software. You can do it at various points the the chain.

Off the top of my head I can’t quite recall the best method in Resolve for you Geoff but I can always take a look after the weekend if you’re still interested.

Being able to process and clamp is definitely the way to deal with those values nicely.

Jack



On Fri, 9 Mar 2018 at 21:27, Justin Johnson <johnson322@...> wrote:
On Fri, Mar 9, 2018 at 10:36 AM, Art Adams <art.cml.only@...> wrote:
>Noise can easily step way outside of gamut in ACES and my guess is that you are seeing some artifacts related the resulting negative values.

Okay, this is good stuff. :) When you say negative values, do you mean numerically negative or simply outside of gamut? And numerically in which axis?


When colors go outside of gamut they produce negative values if they are in a floating point format.   The color primary positions define a point in Yxy space that is 100% saturated.   And if you have, for instance, a red color value that is outside of a color space's triangle, it is going beyond 100% saturation, so green and blue values must go below 0 (or clamp at 0) to compensate.  This results in data loss if those negative values get clamped, and either way, can produce strange artifacts when viewed through a particular output transform. 

 

Would a workflow outside of ACES potentially eliminate the noise in this case, as there would be a different transform to the viewing environment?

Potentially, but I would need to know more specifics about the data flow to say for sure.  Many have been using LMT's to fix artifacts by adjusting the primary values to reign in out-of-gamut values.
 

And wouldn't an ACES workflow take the Canon Cinema Gamut into account so something like this wouldn't happen? This seems like a fairly serious issue, although I guess if it only happens with extreme noise then maybe that's just exceeding the limits of that workflow.

In my expreience, you are likely to see artifacts with ACES on any super saturated values, not just noise.  My belief is that IDTs are supposed to handle this, but their creation is up to the camera vendors and my understanding is that they are usually just matrix transfroms which can push data out-of-gamut.   This is where I yield to those who know more about the subject than I do.   I think the goal of an IDT is to remap the sensor data into the human spectral locus and thus ACES, but again, my understanding of ACES isn't complete.

As you can see below, it seems camera makers are choosing to define their color spaces to encompass or to be drawn near the spectral loci of their sensors, which due to the nature of CMOS sensors is different than the human spectral locus. 

 

--
Cheers,
Jack Jones
CTO, Roundtable Post Production
+44(0)7496 170 481
jack@...
http://www.roundtablepost.co.uk

Colin Elves
 

I think the red noise is a feature of the C200 camera, rather than the ACES workflow. I noticed that the C200 always seem to have incredibly noisy reds in every online ‘test’ or ‘comparison’ video I’ve seen - and looking at my own recent tests the Red chip in the X-rite chart is noticeably noisier than the other cameras tested. There is even noise visible in the 1 stop over exposed test (at ISO 800). The noise is there both in Resolve (applying the standard canon C200 Clog2 to BT709 LUT) and in the Canon Cinema Raw development software preview window. 

I just think it has a lot of noise in the Red channel - one of its quirks. 

Colin Elves
Director of Photography
Berlin/London

Geoff Boyle
 

OK, I’ve gone and tested all the Cinema EOS footage I have, and that’s a lot!

It’s not an ACES issue, it appears when I go straight Resolve mode, managed or not.

The red noise is there in all tests but to nothing like the level in my tests of the C200 last week.

This was a rental company and I suspect that I am at fault because I didn’t do a factory reset which is something I normally do with any rental kit.

I always do a complete reset of any cameras we use in CML tests. If I didn’t I’d be in a mess because sometimes they arrive in a hell of a state, the EVA I had recently was dreadful until I did a complete reset.

I didn’t have this problem on Adult Babies so I looked at what I did differently there.

It’s very simple, instead of cinema gamut I had the camera set to rec 2020. That’s the only difference I can find.

More tests to be done and I’ll post the results when I’ve done them.

 

Cheers

 

Geoff Boyle NSC FBKS

Cinematographer

Zoetermeer

www.gboyle.co.uk

+31 (0) 637 155 076

 

Colin Elves
 

See. This is why I think the C200 is such a weird camera. If it was really ‘Raw’ surely changing gamut in camera shouldn’t make  such a huge difference to the noise profile?

I don’t get why they don’t just call it a 12 bit 444 camera and stop pretending it’s Raw. 

That and I don’t get why anyone would use a camera professionally that won’t let you monitor in the gamma it’s Shooting in. Why no Clog2 monitoring?! 

It’s a very odd beast.

Colin Elves
Director of Photography
London/Berlin.



On 10 Mar 2018, at 07:55, Geoff Boyle <geoff.cml@...> wrote:


I didn’t have this problem on Adult Babies so I looked at what I did differently there.

It’s very simple, instead of cinema gamut I had the camera set to rec 2020. That’s the only difference I can find.

More tests to be done and I’ll post the results when I’ve done them.

 

Art Adams
 

Jim and Justin- thanks very much. That’s great information. It's interesting that this is possible. 

It sounds as if some camera manufacturers are a bit conservative in their transforms. I suspect there must be a reason, or maybe noise is just such a wild card that it's not worth considering beyond a certain point as "you get what you get."

I'm curious about the non oversaturated channels having to "compensate." Why is that. It would make sense if the only way to saturate a channel beyond 100% would be to reduce saturation in the others in impossible ways, but I'm not sure I have that right.

Thanks much.

--
Art Adams
Director of Photography
San Francisco Bay Area

Fahnon
 

I'm late to the party on this one, but talking with someone else who was experiencing noise on the C200, the issue was that they hadn't black balanced the camera.  Not sure if that's happening here, but I find that the C line really needs to black balance or you get noisy images.  Page 53 of the C200 manual says you should BB after every ISO change; after shooting slomo (or changing back from shooting slomo) and even after a factory reset.

@Geoff: I own a C200 and if you don't have the camera anymore I would be happy to help you test.



--
Fahnon Bennett
Filmmaker/Photographer
Brooklyn, New York
323.375.4332