Z7 PDAF banding technical discussion

Started 6 months ago | Discussions
JimKasson
JimKasson Forum Pro • Posts: 24,231
Z7 PDAF banding technical discussion
5

The original thread is about to fill up, so I'm starting a new one to carry on the discussion.

Thanks to all who have participated, and who will do so.

Jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Nikon Z7
If you believe there are incorrect tags, please send us this post using our feedback form.
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Banding evanescence?
1

It's been hard to reproduce the banding consistently. Is there a shot to shot variation?

There are questions about how much what's going on in one side of the image affect banding in the other.

I set up an experiment to shed light on both those things.

I put a light on the right side of an otherwise-pretty-dark scene.

I made 10 exposures with the camera set to ISO 64, the shutter to 1/640, the 24-70mm f/4 at 70 mm and f/4. That was enough exposure to get the light within a stop or two of saturation, but definitely below that.

Then I turned down the light a bit and made 10 more exposures. I did that for 4 sets, then I turned the light off and made 10 exposures, for 50 exposures in all.

Then cropped all the images to the left half, and plotted the mean values of the PDAF rows and the non-PDAF rows.

Flare causes the left-hand values to drop as the light is turned down. But the values for each of the sets of ten exposures are consistent, and the difference between the means of the PDAF and non-PDAF rows is not much affected.

My preliminary conclusions are that the banding is consistently there or not there and that it is not heavily affected by light on the other side of the image. I suspect that its visibility is affected by both the setup and the luck of the draw, though.

Jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
A visual
1

Heres' a 1:1 crop of the left side of the first image in the series, with Exposure set to +5 stops and Shadows set to +100 in ACR Process Version 5.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
A proposed mitigation strategy
2

It may be too early in our understanding of the Z7 banding to start proposing fixes but based on my previous experiments here's what may be a starting place for fixing the banding during raw development.

  • Add x counts to the blue pixels in the PDAF rows of the raw file.
  • counts to the Gr pixels in the PDAF rows of the raw file.
  • Develop normally from then on.

A starting place for x is 1.3 counts. A starting place for x is 0.5 counts. It would be great if that didn't vary by camera model, lighting situation, ISO setting, and phase of the moon.

pippo27, what say you?

Jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
pippo27 Regular Member • Posts: 354
Re: A proposed mitigation strategy

Hi,

JimKasson wrote:

It may be too early in our understanding of the Z7 banding to start proposing fixes but based on my previous experiments here's what may be a starting place for fixing the banding during raw development.

  • Add x counts to the blue pixels in the PDAF rows of the raw file.
  • counts to the Gr pixels in the PDAF rows of the raw file.
  • Develop normally from then on.

A starting place for x is 1.3 counts. A starting place for x is 0.5 counts. It would be great if that didn't vary by camera model, lighting situation, ISO setting, and phase of the moon.

pippo27, what say you?

Challenge accepted If you send me a NEF file with banding (in a format for which there is a publicly-available decoder, i.e. not the new uncompressed one), I'll take a look and see what can be done. Bonus if there is already a mapping of which are the PDAF rows -- otherwise I'll try to figure that out from the banding.

Alberto

PS: I can promise to try, not to succeed of course...

Xasan Contributing Member • Posts: 778
Re: Banding evanescence?

Is it possible that OSPDAF pixels start in the second gain mode, and that is a contributing factor?

Iliah Borg Forum Pro • Posts: 25,444
Re: Banding evanescence?

Xasan wrote:

Is it possible that OSPDAF pixels start in the second gain mode, and that is a contributing factor?

If it is the case, and statistical analysis should show it, the solution may be as simple as adjusting black levels for those pixels involved.

-- hide signature --
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Re: Banding evanescence?

Iliah Borg wrote:

Xasan wrote:

Is it possible that OSPDAF pixels start in the second gain mode, and that is a contributing factor?

It doesn't appear to be so. other than the conversion gain change decreases read noise.

If it is the case, and statistical analysis should show it, the solution may be as simple as adjusting black levels for those pixels involved.

Right. Possibly like this:

https://www.dpreview.com/forums/post/61811777

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Xasan Contributing Member • Posts: 778
Re: Banding evanescence?

Iliah Borg wrote:

Xasan wrote:

Is it possible that OSPDAF pixels start in the second gain mode, and that is a contributing factor?

If it is the case, and statistical analysis should show it, the solution may be as simple as adjusting black levels for those pixels involved.

I guess one can try estimating the gradient of black level along each affected segment. I remember you saying all along that black level in metadata can't be trusted. Problem may be lying not just in black level. A study of FPN close to saturation may provide additional hints to the nature of the effect. I hope NIkon is still using normalization of analog signal before feeding it to S&H and ADC. Additional algorithmic interpolation in firmware is mostly excessive in this case, unless it is very minor. Such interpolation certainly shouldn't smear a pixel, unless they have a bug. On the other hand, even minor adjustments in firmware may change black levels and need different white balance multipliers.

Marianne Oelund Veteran Member • Posts: 7,657
Re: A visual
2

JimKasson wrote:

Heres' a 1:1 crop of the left side of the first image in the series, with Exposure set to +5 stops and Shadows set to +100 in ACR Process Version 5.

Why are the bands slanting upward from left to right, instead of being precisely aligned with the top/bottom image edges?

-- hide signature --

Source credit: Prov 2:6
- Marianne

NowHearThis
NowHearThis Veteran Member • Posts: 3,768
Small favor to ask...
2

JimKasson wrote:

Heres' a 1:1 crop of the left side of the first image in the series, with Exposure set to +5 stops and Shadows set to +100 in ACR Process Version 5.

I’m no doubt in the minority here, I doubt banding will ever be an issue for me.  I don’t push the exposure past +0.33EV hardly ever ( maybe 1 in 1000 shots), I also don’t slide the shadows higher than 80 most of the time and usually I’m somewhere between 30-50.  Given this is my norm, would I see banding?   TRhis is a simplified question and I’m just looking for a simple answer.

-- hide signature --

NHT

 NowHearThis's gear list:NowHearThis's gear list
Olympus PEN-F Panasonic Leica 12-60mm F2.8-4.0 ASPH Olympus M.Zuiko Digital ED 40-150mm 1:4-5.6 R Panasonic Lumix G 42.5mm F1.7 Panasonic Lumix G 25mm F1.7 ASPH
Horshack Veteran Member • Posts: 7,339
Re: A visual
2

Marianne Oelund wrote:

JimKasson wrote:

Heres' a 1:1 crop of the left side of the first image in the series, with Exposure set to +5 stops and Shadows set to +100 in ACR Process Version 5.

Why are the bands slanting upward from left to right, instead of being precisely aligned with the top/bottom image edges?

ACR applying lens correction using the built-in profile embedded in the raw:

https://www.dpreview.com/forums/post/61789941

https://www.dpreview.com/forums/post/61790082

pippo27 Regular Member • Posts: 354
Re: A proposed mitigation strategy
6

Hi,

pippo27, what say you?

Challenge accepted If you send me a NEF file with banding (in a format for which there is a publicly-available decoder, i.e. not the new uncompressed one), I'll take a look and see what can be done. Bonus if there is already a mapping of which are the PDAF rows -- otherwise I'll try to figure that out from the banding.

RawTherapee to the rescue!

Jim's NEF, neutral profile, pushed 8 stops (forget about the colours, RT doesn't have a colour profile for the Z7 yet -- BTW, if you have a colour target and want to contribute, here  are instructions on how to do so):

After one of the PDAF striping correction methods implemeted in RT:

How it is done:

If you want a more technical explanation, the line noise filter does a smoothing of the image in the frequency domain; in this case, this is limited to the PDAF rows only (this requires knowing the pattern in advance, which fortunately is easy), to avoid excessive image quality degradation.

syberman7 Senior Member • Posts: 1,171
Re: A proposed mitigation strategy

pippo27 wrote:

Hi,

pippo27, what say you?

Challenge accepted If you send me a NEF file with banding (in a format for which there is a publicly-available decoder, i.e. not the new uncompressed one), I'll take a look and see what can be done. Bonus if there is already a mapping of which are the PDAF rows -- otherwise I'll try to figure that out from the banding.

RawTherapee to the rescue!

Jim's NEF, neutral profile, pushed 8 stops (forget about the colours, RT doesn't have a colour profile for the Z7 yet -- BTW, if you have a colour target and want to contribute, here are instructions on how to do so):

After one of the PDAF striping correction methods implemeted in RT:

How it is done:

If you want a more technical explanation, the line noise filter does a smoothing of the image in the frequency domain; in this case, this is limited to the PDAF rows only (this requires knowing the pattern in advance, which fortunately is easy), to avoid excessive image quality degradation.

That seems to work very well at eliminating the regular pattern, although it does introduce some artifacting in the image, such as around the jug in the Cuisinart Coffee machine. All the text on the recipe books gets a little softer and less contrasty as well. But certainty the nasty pattern is eliminated which might prove very useful for some images.

 syberman7's gear list:syberman7's gear list
Olympus E-M1 II Nikon Z6 Nikon AF-S Micro-Nikkor 60mm F2.8G ED Nikon AF-S Nikkor 85mm f/1.4G Olympus 12-40mm F2.8 +10 more
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Re: A proposed mitigation strategy
1

pippo27 wrote:

Hi,

pippo27, what say you?

Challenge accepted If you send me a NEF file with banding (in a format for which there is a publicly-available decoder, i.e. not the new uncompressed one), I'll take a look and see what can be done. Bonus if there is already a mapping of which are the PDAF rows -- otherwise I'll try to figure that out from the banding.

RawTherapee to the rescue!

Jim's NEF, neutral profile, pushed 8 stops (forget about the colours, RT doesn't have a colour profile for the Z7 yet -- BTW, if you have a colour target and want to contribute, here are instructions on how to do so):

After one of the PDAF striping correction methods implemeted in RT:

How it is done:

If you want a more technical explanation, the line noise filter does a smoothing of the image in the frequency domain; in this case, this is limited to the PDAF rows only (this requires knowing the pattern in advance, which fortunately is easy), to avoid excessive image quality degradation.

Wow! That was quick! And effective. In real life the affected areas are going to be a lot darker than this, so I think the level of artifacting is entirely acceptable.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Re: Small favor to ask...

NowHearThis wrote:

JimKasson wrote:

Heres' a 1:1 crop of the left side of the first image in the series, with Exposure set to +5 stops and Shadows set to +100 in ACR Process Version 5.

I’m no doubt in the minority here, I doubt banding will ever be an issue for me. I don’t push the exposure past +0.33EV hardly ever ( maybe 1 in 1000 shots), I also don’t slide the shadows higher than 80 most of the time and usually I’m somewhere between 30-50. Given this is my norm, would I see banding? TRhis is a simplified question and I’m just looking for a simple answer.

Probably not, but here's a caveat: you can get hidden pushes as the raw developer compensates for lens light falloff away from the optical axis. Over on the Sony FF MILC forum, we've seen some lenses designed with digital correction in mind need quite a bit of that.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
pippo27 Regular Member • Posts: 354
Re: A proposed mitigation strategy

Hi,

That seems to work very well at eliminating the regular pattern, although it does introduce some artifacting in the image, such as around the jug in the Cuisinart Coffee machine. All the text on the recipe books gets a little softer and less contrasty as well.

Yes, this is a workaround, not a perfect solution. Also, 25 is a bit too aggressive, 10 seems to work just fine and gives less artifacts. But please keep in mind this is an 8-stop push, hopefully not something you do often...

JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Re: A proposed mitigation strategy
1

pippo27 wrote:

Hi,

That seems to work very well at eliminating the regular pattern, although it does introduce some artifacting in the image, such as around the jug in the Cuisinart Coffee machine. All the text on the recipe books gets a little softer and less contrasty as well.

Yes, this is a workaround, not a perfect solution. Also, 25 is a bit too aggressive, 10 seems to work just fine and gives less artifacts. But please keep in mind this is an 8-stop push, hopefully not something you do often...

I'm certainly not complaining!

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
pippo27 Regular Member • Posts: 354
Re: A proposed mitigation strategy
2

Hi,

Wow! That was quick! And effective. In real life the affected areas are going to be a lot darker than this, so I think the level of artifacting is entirely acceptable.

interestingly, although the appearance of the artifact is very different, I was able to fully reuse the code added for the Sony PDAF striping. In fact, I didn't have to add a single line of code, I merely included the info about the Z7 PDAF pattern, that's why it was quick (and that's also why I didn't follow your suggested approach: before implementing something new, I thought I'd check what we already have...

JimKasson
OP JimKasson Forum Pro • Posts: 24,231
Re: A proposed mitigation strategy

pippo27 wrote:

Hi,

Wow! That was quick! And effective. In real life the affected areas are going to be a lot darker than this, so I think the level of artifacting is entirely acceptable.

interestingly, although the appearance of the artifact is very different, I was able to fully reuse the code added for the Sony PDAF striping. In fact, I didn't have to add a single line of code, I merely included the info about the Z7 PDAF pattern, that's why it was quick (and that's also why I didn't follow your suggested approach: before implementing something new, I thought I'd check what we already have...

And it worked! When will the new version be available for download?

jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Keyboard shortcuts:
FForum MMy threads