Topics

Noise reduction in Red’s IPP2

Colin Elves
 

Merry Xmas everyone!

Can anyone confirm whether or not Red’s Image Processing Pipeline incorporates some noise reduction?

I’m assuming it does - otherwise there’d be FPN at high ISOs, I don’t really understand how they can magically improve noise performance from one IPP to the next without it (and also: what would be the point of black shading?) but don’t want to be declaring this as fact without some support!

I’m also curious that the CML tests show they deliver less resolution than the uncompressed 4K from the F55 - so wondering if Noise Reduction is part of that (and/or their compression).

Thanks!

Colin Elves
Curious DP in Berlin.

Paul Curtis
 

On 23 Dec 2017, at 11:12, Colin Elves <colin@...> wrote:
Can anyone confirm whether or not Red’s Image Processing Pipeline incorporates some noise reduction?
AFAIK the official answer is no. But then there are a number of options when debayering the files. So whilst the source RAW files won't have any, depending on how the image is generated you have various choices there. From sharpness during debayer, to DEB noise reduction (subtle but can work wonders for colour noise) to 'normal' sharpening. So when you see results of tests unless you know the various settings they're not necessarily that meaningful.

Again my understanding of IPP2 is more about colourspace conversions and tone mapping. It does an amazing job, first camera i've had where i can point and be happy with the image as is.

FPN i presume is dealt with separately.

You know where to get me if you want any tests or anything doing...

cheers
Paul

Paul Curtis, VFX & Post | Canterbury, UK

Mark Kenfield
 

The F55/F65/F5 don't do uncompressed raw. They offer 3:1 compression.


Cheers,

Mark Kenfield
Cinematographer
0400 044 500

I’m also curious that the CML tests show they deliver less resolution than the uncompressed 4K from the F55 - so wondering if Noise Reduction is part of that (and/or their compression).

Art Adams
 


Can anyone confirm whether or not Red’s Image Processing Pipeline incorporates some noise reduction?

A rental house I work with showed me some processed RED Helium footage and there was clearly noise reduction happening. I don't know what their settings were, but they claimed they'd not added any noise reduction at all. Their reference (that they use for all their cameras) is a piece of very black textured cloth. Other cameras rendered it with some texture, while Helium made it appear a completely smooth yet rich black.

A while back I remember seeing a test online where the camera was rated at different ISOs and as they increased the image became green. It seemed that the camera was pulling green noise out of the shadows as the ISO increased. I was told this happened because the camera wasn't white balanced with each ISO change, which is apparently the recommended procedure. When that's done the green noise cleans up nicely.

The only reason white balance would make a difference is if there was some sort of noise reduction happening in the processing chain, as that's standard procedure for every other camera as they all do some noise cleanup based on what the white balance is. 3200K? Clean up the blue channel. 5500K? Clean up the red channel. This generally has a positive effect on color overall and allows for higher ISOs.

The only thing that's unusual about RED is that I don't believe they've had noise reduction kick in automatically before, if that is indeed the case in the footage described above.

My suspicion is that they're trying to cram so many photo sites onto a small 8K sensor that the reduced well size means a lower signal-to-noise ratio. This is something every manufacturer struggles with, and why others are using larger sensors to achieve higher resolution (photo sites can be bigger, so signal-to-noise increases). We're not at the point where you can capture that much resolution off that small a sensor without paying some sort of price, although that's true of every camera: tradeoffs are the name of the game in sensor and camera design.

>I’m also curious that the CML tests show they deliver less resolution than the uncompressed 4K from the F55 - so wondering if Noise Reduction is part of that (and/or their compression).

That sounds wrong. Unless the tests were shot at a crazy high ISO where noise reduction is working across the entire image and not just in the darker tones, I'd suspect a post pipeline problem. The rule of thumb is to knock off 20% of the resolution due to the de-mosaicing process, so that means that an 8K sensor is delivering less than 3.2K resolution even after downsampling to 4K.

I'd think we'd be hearing a lot more complaints if that were true.

--
Art Adams
Director of Photography
San Francisco Bay Area

Ido Karilla
 

In camera recording there is no de noise or detail going on. While debyering you can choose your flavor. 

Sent from my iPhone

On 23 Dec 2017, at 14:28, Mark Kenfield <mark@...> wrote:

The F55/F65/F5 don't do uncompressed raw. They offer 3:1 compression.


Cheers,

Mark Kenfield
Cinematographer
0400 044 500

I’m also curious that the CML tests show they deliver less resolution than the uncompressed 4K from the F55 - so wondering if Noise Reduction is part of that (and/or their compression).

Ido Karilla
 

I think you are making too menu assumptions. With RED it is mostly knowing when to use what. 

The camera does not do any noise reduction from sensor to raw (r3d). 

While debayering  a world of options come to force. Including de noise, detail, drx witch take information from healthy Chanel’s to the week ones.....



Sent from my iPhone

On 23 Dec 2017, at 19:43, Art Adams <art.cml.only@...> wrote:


Can anyone confirm whether or not Red’s Image Processing Pipeline incorporates some noise reduction?

A rental house I work with showed me some processed RED Helium footage and there was clearly noise reduction happening. I don't know what their settings were, but they claimed they'd not added any noise reduction at all. Their reference (that they use for all their cameras) is a piece of very black textured cloth. Other cameras rendered it with some texture, while Helium made it appear a completely smooth yet rich black.

A while back I remember seeing a test online where the camera was rated at different ISOs and as they increased the image became green. It seemed that the camera was pulling green noise out of the shadows as the ISO increased. I was told this happened because the camera wasn't white balanced with each ISO change, which is apparently the recommended procedure. When that's done the green noise cleans up nicely.

The only reason white balance would make a difference is if there was some sort of noise reduction happening in the processing chain, as that's standard procedure for every other camera as they all do some noise cleanup based on what the white balance is. 3200K? Clean up the blue channel. 5500K? Clean up the red channel. This generally has a positive effect on color overall and allows for higher ISOs.

The only thing that's unusual about RED is that I don't believe they've had noise reduction kick in automatically before, if that is indeed the case in the footage described above.

My suspicion is that they're trying to cram so many photo sites onto a small 8K sensor that the reduced well size means a lower signal-to-noise ratio. This is something every manufacturer struggles with, and why others are using larger sensors to achieve higher resolution (photo sites can be bigger, so signal-to-noise increases). We're not at the point where you can capture that much resolution off that small a sensor without paying some sort of price, although that's true of every camera: tradeoffs are the name of the game in sensor and camera design.

>I’m also curious that the CML tests show they deliver less resolution than the uncompressed 4K from the F55 - so wondering if Noise Reduction is part of that (and/or their compression).

That sounds wrong. Unless the tests were shot at a crazy high ISO where noise reduction is working across the entire image and not just in the darker tones, I'd suspect a post pipeline problem. The rule of thumb is to knock off 20% of the resolution due to the de-mosaicing process, so that means that an 8K sensor is delivering less than 3.2K resolution even after downsampling to 4K.

I'd think we'd be hearing a lot more complaints if that were true.

--
Art Adams
Director of Photography
San Francisco Bay Area

Philip Holland
 

If we are talking IPP2, it features a new demosaic algorithm (though that actually should have more detail than before) and beyond that if you look in REDCINE-X PRO you can see there's a denoise slider as well as DEB, which will effect chromatic noise.    I would investigate if any of those settings have been used when rendering out the clips.  I would also look at how the material was exported or viewed as well as which REDCODE RAW Compression Ratio was used.

There shouldn't be a "smoothed out" heavy NR look though in any of these cases
.  

Without seeing an image however, it's tricky to really tell.  If they have an .R3D snapshot or something I could dig into it.
 

Art Adams
 

If you'll look back at my post, you'll see that I said the footage was processed, but didn't say how.

I also pointed out that I'd been told that white balancing seemed to make a difference, and noted why it typically makes that difference.

I didn't say anything about noise processing happening in the camera. I did say that the people who processed the footage didn't do anything that they believed would result in noise reduction, but who knows? This is easy enough for anyone to test.

--
Art Adams
Director of Photography
San Francisco Bay Area

Art Adams
 

>If they have an .R3D snapshot or something I could dig into it.

After the New Year I'll go see if I can go in and make my own. That's the only way to really know, at least for me—although I don't know all the details of that pipeline. I haven't processed RED footage myself in a long time.

--
Art Adams
Director of Photography
San Francisco Bay Area

Colin Elves
 


Ah... glad I managed to start a discussion amongst the ‘gods’ of cinematography with my burnt offering and pleas for assistance on the eve of Saturnalia...

The resolution observation is purely from the last round of CML tests here: https://www.cinematography.net/CE-2017%20STILLS.html

Where the ‘8K’ Red Helium clearly delivers less resolution (and more aliasing) than the F55 (and yes, thanks for correcting me - Sony Raw is indeed lightly compressed. Much lighter than is possible on Helium Cameras AFAIK).

I’m assuming no NR was involved, but Geoff is probably best placed to confirm that. 

It might well be the case that Red previously did not include NR in their IPP - but this does not mean they don’t do so now. It might also simply be that they have it switched on as default in the processing software - and one has to go in and turn it off.

Currently I’m going on pure logic here: The simplest explanation for the Helium delivering less resolution and lower noise using sensor sites much much much smaller than other manufacturers can manage, is that they use some form of noise reduction to deliver an acceptable image (whilst still being able to market high resolution capture) and this reduces the effective resolution as well as reducing the noise. 

I do sort of wonder if the also increase the NR at higher ISOs and/or apply it selectively to different bits of the curve. I guess you could check this with a simple resolution chart by Shooting it at various exposure levels and seeing if there is an identifiable drop in resolution. 

The other way might be to process the same resolution chart using both IPP2 and IPP1 - and seeing if there is a drop in resolution (of that’s possible these days).

Frohe Weihnachten! 

Colin Elves, 
DP/troublemaker 
Berlin für Weihnacten.

Art Adams
 


Currently I’m going on pure logic here: The simplest explanation for the Helium delivering less resolution and lower noise using sensor sites much much much smaller than other manufacturers can manage, is that they use some form of noise reduction to deliver an acceptable image (whilst still being able to market high resolution capture) and this reduces the effective resolution as well as reducing the noise.

I'm going to disagree. I'd have to take a look at what you're talking about, but if the Helium shows less resolution than the Sony F55 then that's not noise reduction, that's a failure in the test pipeline. Even with noise reduction, that shouldn't happen. I'd expect to see loss of resolution in the darkest shadows near black, but nothing like what you're describing. Worst case scenario, the Helium footage could have been recorded with insanely high compression, but I don't see Geoff doing that.

Do the test notes say anything about how the 8K image was brought down to a more manageable size like 4K?

How are you making your comparison? Looking at a test chart image? There's only one 4K test chart that I know of. Most are 2K. I'm curious how they would look after capture on an 8K camera and reduced in size for comparison. I haven't done that yet.

Can you post links to what you're looking at? I'm in the throes of managing family holiday expectations so I can't go on a hunt now, but I might be able to take a quick look tonight.

I've not heard anyone else complain about the Helium's resolution. The noise reduction test I mentioned previously showed nothing like what you describe.
 
--
Art Adams
Director of Photography
San Francisco Bay Area

Mark Weingartner, ASC
 

 Useful input posted by Phil Holland who will remember to sign next time:-)  (says the very occasional listmum)

On 23Dec, 2017, at 12:30 50, phil@... wrote:

If we are talking IPP2, it features a new demosaic algorithm (though that actually should have more detail than before) and beyond that if you look in REDCINE-X PRO you can see there's a denoise slider as well as DEB, which will effect chromatic noise.    I would investigate if any of those settings have been used when rendering out the clips.  I would also look at how the material was exported or viewed as well as which REDCODE RAW Compression Ratio was used.

There shouldn't be a "smoothed out" heavy NR look though in any of these cases
.  

Without seeing an image however, it's tricky to really tell.  If they have an .R3D snapshot or something I could dig into it.

Colin Elves
 


Do the test notes say anything about how the 8K image was brought down to a more manageable size like 4K?
Hi Art!

I’ve got 12 with us over for Xmas, so I feel your pain!

The link was buried in my lease email, so here it is at the top:


The notes say the captures are 1:1 so that would mean no scaling was involved from 8K. 

If the the chart is only 4K then it pretty much confirms what I’ve been suggesting: that the Helium seems to deliver less than 4K.

 But I’m also not too sure if that matters relative to the lenses: make the chart smaller in the frame and you’ll need more resolution to resolve the finer detail (assuming the lenses are capable of it).

Anyway. Something is going on to kill the resolution and create aliasing. And it’s not just the charts. You see it in a lot of comparisons. Helium is just less sharp. 

‘Tis a mystery!

Cheers,

Colin Elves
DP/B&B operator, Berlin

Geoff Boyle
 

All the CML tests were shot at the recommended ISO’s and the post workflow for RED was the one recommended by Graeme Nattress.

I don’t just chuck this stuff up there, I check with the manufacturers and the post software companies.

If you’r din any doubt then use the EXR’s as these were all produced with the manufacturers software with the exception of the Varicams as there isn’t any manufacturers software for them!


Cheers
Geoff Boyle
Cinematographer
EU Based
www.gboyle.co.uk
+31 (0) 637 155 076



On 23 Dec 2017, at 18:43, Art Adams <art.cml.only@...> wrote:

That sounds wrong. Unless the tests were shot at a crazy high ISO where noise reduction is working across the entire image and not just in the darker tones, I'd suspect a post pipeline problem. 

Geoff Boyle
 

I’m pretty sure I said we shot at 7K to keep the image size as close to S35 as we could and that this was our approach to every camera.


Cheers
Geoff Boyle
Cinematographer
EU Based
www.gboyle.co.uk
+31 (0) 637 155 076



On 24 Dec 2017, at 07:25, Colin Elves <colin@...> wrote:

The notes say the captures are 1:1 so that would mean no scaling was involved from 8K. 

If the the chart is only 4K then it pretty much confirms what I’ve been suggesting: that the Helium seems to deliver less than 4K.

Paul Curtis
 

On 23 Dec 2017, at 23:43, Art Adams <art.cml.only@...> wrote:
I'm going to disagree. I'd have to take a look at what you're talking about, but if the Helium shows less resolution than the Sony F55 then that's not noise reduction, that's a failure in the test pipeline. Even with noise reduction, that shouldn't happen. I'd
Think failure of test pipeline is a bit strong(!) but the one thing about Red is deferring many decisions until post time and there are many different routes to take. Given the same R3D you can make something soft or sharp very easily. I don't know what settings Geoff was recommended, but there's a degree of personal choice and look in there. Having come from what i now perceive as overly sharpened images from previous cameras i prefer the soft smoothness of the Red image with no sharpening.

Also the debayer algorithm itself can have a huge impact on the detail and sharpness of the image as you know, yet for the most part we just accept and treat all the various cameras the same but actually that process has as much influence over the end result as anything else in the chain. If you have DNG files which are more open, then you can try for yourself and push the same image data through many many different debayers and see how different they can be. Of course most times we can't change the debayer but when comparing cameras it's nice to remember that there is an extra step of difference between them.

I assume in the tests being referred to that the lens was the same in all.

I think Geoff has made the most comprehensive and thoughtful tests, but like all tests you really should do your own when it counts.

cheers
Paul

Paul Curtis, VFX & Post | Canterbury, UK

Geoff Boyle
 

Again, thats in the pages on the web, all PL mount were the same lens, the EF mounted cameras had a single lens but it was different.

Cheers
Geoff Boyle
Cinematographer
EU Based
www.gboyle.co.uk
+31 (0) 637 155 076



On 24 Dec 2017, at 10:28, Paul Curtis <paul@...> wrote:

I assume in the tests being referred to that the lens was the same in all.

Colin Elves
 

oh. Maybe. I’m not sure that makes it much better though. It’s still less than advertised!

Colin Elves
DP, Berlin



On 24 Dec 2017, at 08:53, Geoff Boyle <geoff.cml@...> wrote:

I’m pretty sure I said we shot at 7K to keep the image size as close to S35 as we could and that this was our approach to every camera.


Rakesh Malik
 

It doesn't jive with my experience using a Helium. I've been able to extract print quality frames from motion footage, as well as make some rather nice stills with a Helium camera. 

IPP2 doesn't have any noise reduction applied by default. If you want it, you have to turn it on. 

I'm finding that when everything's executed well, I can get pretty much the same resolution and detail as with my A7R, though the Helium has nicer color before grading when using IPP2.

-----------------------------

On Sat, Dec 23, 2017 at 1:01 PM, Colin Elves <colin@...> wrote:

Ah... glad I managed to start a discussion amongst the ‘gods’ of cinematography with my burnt offering and pleas for assistance on the eve of Saturnalia...

The resolution observation is purely from the last round of CML tests here: https://www.cinematography.net/CE-2017%20STILLS.html

Where the ‘8K’ Red Helium clearly delivers less resolution (and more aliasing) than the F55 (and yes, thanks for correcting me - Sony Raw is indeed lightly compressed. Much lighter than is possible on Helium Cameras AFAIK).

I’m assuming no NR was involved, but Geoff is probably best placed to confirm that. 

It might well be the case that Red previously did not include NR in their IPP - but this does not mean they don’t do so now. It might also simply be that they have it switched on as default in the processing software - and one has to go in and turn it off.

Currently I’m going on pure logic here: The simplest explanation for the Helium delivering less resolution and lower noise using sensor sites much much much smaller than other manufacturers can manage, is that they use some form of noise reduction to deliver an acceptable image (whilst still being able to market high resolution capture) and this reduces the effective resolution as well as reducing the noise. 

I do sort of wonder if the also increase the NR at higher ISOs and/or apply it selectively to different bits of the curve. I guess you could check this with a simple resolution chart by Shooting it at various exposure levels and seeing if there is an identifiable drop in resolution. 

The other way might be to process the same resolution chart using both IPP2 and IPP1 - and seeing if there is a drop in resolution (of that’s possible these days).

Frohe Weihnachten! 

Colin Elves, 
DP/troublemaker 
Berlin für Weihnacten.

Colin Elves
 


I'm finding that when everything's executed well, I can get pretty much the same resolution and detail as with my A7R, though the Helium has nicer color before grading when using IPP2.

Ah, now that’s interesting! Have you done a direct comparison? Same lens/stop/subject (ideally a focus chart)? 

I’m not a 100% sure if having the option NR settings in the Raw control panel is quite the same as there being no other noise reduction elsewhere in the image chain. 

I’m assuming there’s already a separate process for dealing with Fixed Pattern Noise. So there’s no reason why there might not be others. 

Colin Elves
DP/contrarian
Germany

Previous Topic Next Topic