URSA 12K banding

Can someone ask them to join this debate?

I'll try to contact them after NAB is over.

Thank you Robert!

I thought diffraction steps in at higher f-stops (11 and up).

Rayleigh's Criterion is a mathematical way of putting a number on the smallest resolvable detail when an imaging system is limited by diffraction. For 4K or HD, F/8-F/11 could be okay. The 12K sensor pixel pitch is only 2.2μm. It can't resolve 12K fully if the smallest possible light spot falls across more than 2-3 pixels. Diffraction sets a resolvable limit.

Rayleigh's Criterion:

θ=1.22λ/D

Where θ is in radians, λ is wavelength of green light; .55μm , D is pixel pitch; 2.2μm.

The diameter of the airy disk is approximately 2-3x the pixel pitch. This is because the pixel pitch corresponds to the spacing between the center of two adjacent pixels, the smallest resolvable detail, and the diameter would cover at least two pixels. F/stop is 1/2 the inverse of numerical aperture (NA), because θ is in radians.

We write this as the smallest resolvable f/stop:

f/stop = 2D/2.44λ

That is, two pixels covered and 2 radians, make 1 numerical aperture, the inverse of which is the f/stop.

For the UMP12K, that's 2 x 2.2 / 2.44 x .55 = f/3.3
For the Cine12K, that's 2 x 2.9 / 2.44 x .55 = f/4.3
For the Pocket 6K that's 2 x 3.8/ 2.44 x .55 = f/5.7
For the UMP4.6G2 that's 2 x 5.5/ 2.44 x. 55 = f/8.2
For the BMCC6K, that's 2 x 6.0 / 2.44 x .55 = f/8.9 (Winner, Chicken Dinner)

If we accept 3 pixels covered by the airy disk instead of just two, we can stop down to get:

For the UMP12K, that's 3 x 2.2 / 2.44 x .55 = f/4.9
For the Cine12K, that's 3 x 2.9 / 2.44 x .55 = f/6.5
For the Pocket 6K that's 3 x 3.8/ 2.44 x .55 = f/8.5
For the UMP4.6G2 that's 3 x 5.5/ 2.44 x. 55 = f/f12.3
For the BMCC6K, that's 3 x 6.0 / 2.44 x .55 = f/13.4

As you can see, diffraction sets the limit on the smallest physical detail. As the pixel pitch gets more densely packed, the sensor will out-resolve the lens at smaller f/stops openings.

...new example...
Constant Quality - Q3 - Zeiss Lens - 8k....
2nd one is isolated red channel. Pretty visible artifacts.
1st one is full spectrum. Artifacts most visible at lower right parts of sky.

Puzzled. This is not suppose to happen...

This not suppose to happen...

You’re not alone. I just noticed the same thing happens in the Braw clip 2, the backlit clip on the balcony.
https://downloads.blackmagicdesign.com/ ... 1.braw.zip

Image 21-04-2024 at 10.42.jpeg (231.66 KiB) Viewed 2766 times

In a “normal grade” it’s not visible, but nonetheless it’s potentially there.

Can't open that BRAW file in Resolve. Not sure why...

Can't open that BRAW file in Resolve. Not sure why...

You need R19 beta.

Okay, I have it on another comp.
What cam is that?

Can only be the new one.

Okay, I have it on another comp.
What cam is that?

The clip was shot to showcase the new Ursa Cine 12K.

It's there on the 12K and not there on the 4.6K.

I do have a much harder time to reproduce it than you it seems, though. Davinci 19 (Mac).

Bildschirmfoto 2024-04-21 um 17.36.jpeg (259.2 KiB) Viewed 2625 times

Bildschirmfoto 2024-04-21 um 17.36.10.jpg (170.7 KiB) Viewed 2625 times

Okay, I really think someone from Blackmagic should address this issue!!

I think you’ll have more luck going through support than trying to work through this on a forum. Have you tried that path?

JB

Hey John!
Support. like BMD service or...? If it's service they will charge me nicely for this even though camera was 2 times there already for more than half a year. So much of their service.
If there were justice, they would send me a new camera...

I have contacted Hook and pointed him to this thread. Hopefully he can chime in and shed some light onto this issue.

This not suppose to happen...

You’re not alone. I just noticed the same thing happens in the Braw clip 2, the backlit clip on the balcony.
https://downloads.blackmagicdesign.com/ ... 1.braw.zip

Image 21-04-2024 at 10.42.jpeg

In a “normal grade” it’s not visible, but nonetheless it’s potentially there.

Yup!
I can see it in red channel. Is that normal??

Yup!
I can see it in red channel. Is that normal??

From what I get, reading this forum over the years, skies can break your expectations.
The science behind it I leave to Hook. He gave a good explanation some years back that might be related.
https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=98464&start=150#p554170

I have no clue if something similar is happening here.

But here is what I would do. Test it.
Four takes @ 12K: iso 200, iso 1000, iso 1250 iso 2500 (with nd compensation). Compression: uncompressed
For each take focus@ infinity and focus @ the houses.

Result: banding send your files over to Blackmagic and hope for the best.
Result: no banding, try again and now with compression.

The focus point might be of importance as focus plays part in how much compression is used.

Clip 2 from Blackmagic showcased an anamorphic flare. Judging the behind the scenes, the same iso settings were used. Difference being Q3 to your 3:1.

The Q3 is a better codec as it should allocated more room to the parts that need it. However the clip was shot @ iso 800 and the sky was out of focus, so the sky might have been squished to band.

Focus has never been a part of the compression algorithm before BRAW as far as I know and might be part of the problem, maybe…

Could this banding issues be due to different sensor technology in 12Ks - old S35 and new FF and how they handle .braw compression?

I checked our footage of clear blue skies from UMPG2 and P6KP and they are all banding free. Our own initial tests for compression artifacts were done in a backlit forest, which I think is the hardest scene for compression. At 12:1 constant we found some artifacts in details, while at 8:1 it completely held up and that's why we go with Q5 almost exclusively. Q5 on our cameras is 20:1 - 7:1.

CINE 12K backlit sample clip with blue sky on top is filmed in Q3 which for this new camera is 30:1 - 12:1.. This might be the reason for banding on the sky?

Mario I am really sorry to hear that you have so many issues with your unit, this really sucks and I hope you have this resolved ASAP.

When we had issues with our G2 we were really lucky with BMD service sending us a new unit so our story with them is a very pleasant experience.

Could this banding issues be due to different sensor technology in 12Ks - old S35 and new FF and how they handle .braw compression?

I checked our footage of clear blue skies from UMPG2 and P6KP and they are all banding free. Our own initial tests for compression artifacts were done in a backlit forest, which I think is the hardest scene for compression. At 12:1 constant we found some artifacts in details, while at 8:1 it completely held up and that's why we go with Q5 almost exclusively. Q5 on our cameras is 20:1 - 7:1.

CINE 12K backlit sample clip with blue sky on top is filmed in Q3 which for this new camera is 30:1 - 12:1.. This might be the reason for banding on the sky?

Mario I am really sorry to hear that you have so many issues with your unit, this really sucks and I hope you have this resolved ASAP.

When we had issues with our G2 we were really lucky with BMD service sending us a new unit so our story with them is a very pleasant experience.

Hey Blaz! You're from Slovenia?
Well, I tested 12k and it's the only camera displaying such artefacts. I tried my UMP 4.6k Pro (see beginning of this thread) and it held up perfectly (same settings as 12k). I also did test my Pocket 6KPro., you guessed right, completely clean skies. It might be some weird thing going on with new 12k sensors they make....

Yup!
I can see it in red channel. Is that normal??

From what I get, reading this forum over the years, skies can break your expectations.
The science behind it I leave to Hook. He gave a good explanation some years back that might be related.
https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=98464&start=150#p554170

I have no clue if something similar is happening here.

But here is what I would do. Test it.
Four takes @ 12K: iso 200, iso 1000, iso 1250 iso 2500. Compression: uncompressed
For each take focus@ infinity and focus @ the houses.

Result: banding send your files over to Blackmagic and hope for the best.
Result: no banding, try again and now with compression.

The focus point might be of importance as focus plays part in how much compression is used.

Clip 2 from Blackmagic showcased an anamorphic flare. Judging the behind the scenes, the same iso settings were used. Difference being Q3 to your 3:1.

The 3:1 is a better codec as it should allocated more room to the parts that need it. However the clip was shot @ iso 800 and the sky was out of focus, so the sky might have been squished to band.

Focus has never been a part of the compression algorithm before BRAW as far as I know and might be part of the problem, maybe…

Hey Bunk!

I did send the issue to BMD contact here, so we'll see what will they say....

Yes Mario good luck with resolving the problem!

Yes Mario good luck with resolving the problem!

Hello neighbor!

Regards from Bregana!

Hi Mario,

This is my two cents, please take with a grain of salt. Just an opinion…

I believe what you are seeing is the limits of the Braw codec based on the compression setting utilized.

Braw is a “partially baked” implementation of raw, unlike previous versions of full raw. It does a partial de-mosiac in-camera which bakes in certain parameters and makes it susceptible to compression artifacts if recorded data rate is too low. With heavily grading in this case, you’re basically “breaking the codec”. Adding excessive saturation to the sky and stacking LUTs (in an area with limited data capture) exposes the artifacts (showing the limitations of chosen data-compression rate). The Blackmagic sensor captures a lot of detail, but by cramming so much resolution into smaller & smaller file sizes, some information invariably suffers.

If this was shot uncompressed I would highly doubt there would be any stair-stepping or aliasing in the sky whatsoever. Since the sky is a mostly continuous blue color, BM’s compression algorithms take a heavier approach in reducing “sky” data, while applying more bits to other detailed areas of the scene. It’s probably a perceptual thing to allocate bits where needed most.

I dropped your clips into an ACES timeline (to get to the least adulterated version of the shot) …and like previous posts mentioned the color aliasing is baked in. On capture, maybe it’s best to increase data rate and shoot Q3 or Q0?
In post-production, maybe the best way to minimize artifacts is to use ACES or process/grade/output in a space with a larger gamut like rec2020?

In the images below, I also see chromatic aberration in the corners of the frame. The color fringing on the umbrella pole is from the lens itself. Hope this is helpful.

Best,

Manuel

Still 2024-04-22 134210_1.5.1.jpeg (502.12 KiB) Viewed 1211 times

Still 2024-04-22 134210_1.3.4.jpeg (805.59 KiB) Viewed 1211 times

Still 2024-04-22 134210_1.5.1.jpeg (502.12 KiB) Viewed 1211 times

Hey John!
Support. like BMD service or...? If it's service they will charge me nicely for this even though camera was 2 times there already for more than half a year. So much of their service.
If there were justice, they would send me a new camera...

No not sending in your camera.

Raising it as a support ticket with the files.

To be honest these are pretty extreme grades that mean you're bringing out these artefacts. My instincts are that you're at the limits of what the compression can handle. The files don't go on forever, remember these are very compressed. Blue Sky can actually be difficult to compress and then you're doing very extreme grades that you probably wouldn't do in a normal situation for a client or delivery.

I would ask through BMD service their opinion.

JB

Hey John!

Correction, extreme grades were those examples at the beginning. The last ones I uploaded are completely normal grades. Even the sat was raised just a tad. It is clearly visible in red channel and in full spectrum is also visible. I don’t want my footage to have “waves” in the sky.
And at the end it doesn’t explain the fact why this issue doesn’t happen with UMP 4.6k Pro and Pocket 6k Pro….even with extreme grades. Sky is smooth as butter!

The Q3 is a better codec as it should allocated more room to the parts that need it. However the clip was shot @ iso 800 and the sky was out of focus, so the sky might have been squished to band.

Focus has never been a part of the compression algorithm before BRAW as far as I know and might be part of the problem, maybe…

Focus its not part of compression per se, but low-contrast of small detail is. Any I-frame-only compression I know about is using the fact that our eye/brain system can't discern small detail at low contrast. So, these are the areas that will get less data allocated than those with high contrast and small detail – which are the ones in focus. Could it be that the Q3 algorithm is just too nice to high detail and starving low detail areas of data, while keeping the overall data rate low?
It might be interesting to compare the highest rate of fixed compression under exactly identical conditions.

Focus its not part of compression per se, but low-contrast of small detail is. Any I-frame-only compression I know about is using the fact that our eye/brain system can't discern small detail at low contrast. So, these are the areas that will get less data allocated than those with high contrast and small detail – which are the ones in focus. Could it be that the Q3 algorithm is just too nice to high detail and starving low detail areas of data, while keeping the overall data rate low?
It might be interesting to compare the highest rate of fixed compression under exactly identical conditions.

You formulated it way better than I did, but that’s the gist of it in combination with the RGBW sensor, as it seems to be more sensitive for it than with the other cameras.

Could it be that the Q3 algorithm is just too nice to high detail and starving low detail areas of data, while keeping the overall data rate low?

It's an interesting suggestion, however it doesn't explain why the same variable codec on the 4.6 UMPG2, may exhibit no banding, notwithstanding the much higher photosite count. I don't think it is classic bit starved banding, such as we still see on 8 bit SDR UHD streaming content sometimes; which tends to have noticeable straight edges to the bands; and out of focus uniform blur areas are often classic betrayers of this. This kind of banding will tend to move around too.

Although it follows the 'banding' of banding, as it were, it maybe just possibly some kind of FPN interference; perhaps caused by an imperfect OLPF (or any other lens/filtration conflict?); even possibly one not sitting right. It might be worth shooting the same scene in full sensor 4K too, to see if it's still there and if there is access to a Video Assist, 4K Prores/DNxHR might be worth seeing too; more to eliminate or confirm any Braw codec 'issues' than suggest my theory.

It should be pointed out too, some of us are viewing this more prominently than others. These kind artefacts can only be properly determined on a true 10bit reference monitor, to eliminate viewing prejudice. Some claiming to be are only 8+2 FRC (This includes most non-oled consumer HDR TVs; both should be OK too but may cloud things; if you'll excuse the pun) and laptop/pc monitors usually 8bit or 6bit dithered. On my reference monitor it is a very slight thing and you really have to push the grade and zoom in to see it under normal conditions

As JB says though, hopefully it's best if BMD are able to have a proper look, rather than any further speculation on our part.

Focus is definitely affecting the compression.

When you have a sharp image with high frequency detail (leaves for example) then yes the algorithm will allocate more data use a higher number.

If the shot has less high frequency detail and is soft then it allocates less detail.

I wasn’t following from the start but has the OP tried with a different higher fixed rate?

JB

Doesn't seem so, that's why I suggested it.

I had something similar happen on my Pocket 6k on a clip shot in Braw. In my case it was an imperfect wall. It seems like this happens on colors that are very close in detail.

Focus is definitely affecting the compression.

When you have a sharp image with high frequency detail (leaves for example) then yes the algorithm will allocate more data use a higher number.

If the shot has less high frequency detail and is soft then it allocates less detail.

I wasn’t following from the start but has the OP tried with a different higher fixed rate?

JB

I will test again with other compression fixed rate….when clouds go away. Even though I think those samples at the beginning were different compressions. Must check.

Could it be that the Q3 algorithm is just too nice to high detail and starving low detail areas of data, while keeping the overall data rate low?

It's an interesting suggestion, however it doesn't explain why the same variable codec on the 4.6 UMPG2, may exhibit no banding, notwithstanding the much higher photosite count. I don't think it is classic bit starved banding, such as we still see on 8 bit SDR UHD streaming content sometimes; which tends to have noticeable straight edges to the bands; and out of focus uniform blur areas are often classic betrayers of this. This kind of banding will tend to move around too.

Although it follows the 'banding' of banding, as it were, it maybe just possibly some kind of FPN interference; perhaps caused by an imperfect OLPF (or any other lens/filtration conflict?); even possibly one not sitting right. It might be worth shooting the same scene in full sensor 4K too, to see if it's still there and if there is access to a Video Assist, 4K Prores/DNxHR might be worth seeing too; more to eliminate or confirm any Braw codec 'issues' than suggest my theory.

Wouldn’t say it’s OLPF since the same one (Rawlite) is on 4.6k and there’s no banding.
I could try to test with Shogun and see if it’s a BRAW thingie.

But he shot 12K at 5:1 - this surely shouldn't cause banding? Especially when the 4.6K shows no banding at all when shooting that same sky?

Wouldn’t say it’s OLPF since the same one (Rawlite) is on 4.6k and there’s no banding.
I could try to test with Shogun and see if it’s a BRAW thingie.

That doesn't necessarily rule it out Mario; different target sensor resolutions - I thought you had a 12K OLPF (i.e. the BMD designed one). It maybe something since they made changes to Braw to work with that filter. The Rawlites are great and top notch - I have one on my Ursa Broadcast G2 but obviously Hans has no control over software. It was just a suggestion too, amongst any number of possibilities and I'm starting to sound like a broken record on them. Have you got anywhere with contacting BMD themselves?

We’re well into speculation now. I think the best action is to go direct to BMD.

JB

The UM12k footage shows a lot of color noise - especially compared to the UM46k footage. So it could be possible that BRAW encoding exhausted its available bitrate contingent and then had to compress the low detail sky parts too much.

I've now looked at the UM12k ARIA footage John Brawley provided on his blog - the sky shows no banding
-> 12KAriaDUSK_21mm_no_filter.braw shot at Q3

Wouldn’t say it’s OLPF since the same one (Rawlite) is on 4.6k and there’s no banding.
I could try to test with Shogun and see if it’s a BRAW thingie.

That doesn't necessarily rule it out Mario; different target sensor resolutions - I thought you had a 12K OLPF (i.e. the BMD designed one). It maybe something since they made changes to Braw to work with that filter. The Rawlites are great and top notch - I have one on my Ursa Broadcast G2 but obviously Hans has no control over software. It was just a suggestion too, amongst any number of possibilities and I'm starting to sound like a broken record on them. Have you got anywhere with contacting BMD themselves?

Hey Steve, wrote to BM support and still nothing. Been 2 day now…

But he shot 12K at 5:1 - this surely shouldn't cause banding? Especially when the 4.6K shows no banding at all when shooting that same sky?

Agreed!!

I've now looked at the UM12k ARIA footage John Brawley provided on his blog - the sky shows no banding
-> 12KAriaDUSK_21mm_no_filter.braw shot at Q3

Hi Robert,
Could you provide a link to the file.

Thanks

I've now looked at the UM12k ARIA footage John Brawley provided on his blog - the sky shows no banding
-> 12KAriaDUSK_21mm_no_filter.braw shot at Q3

Hi Robert,
Could you provide a link to the file.

Thanks

You can find it on John's blog:
https://johnbrawley.wordpress.com/2022/ ... ter-tests/

Direkt link to the files download area John provided:
https://bit.ly/3TyIIIW

I've now looked at the UM12k ARIA footage John Brawley provided on his blog - the sky shows no banding
-> 12KAriaDUSK_21mm_no_filter.braw shot at Q3

Hi Robert,
Could you provide a link to the file.

Thanks

Thanks Robert!
I checked those Aria clips from John and I think it is not entirely identical situation. In his clips we can see less
sky than in my clips. Also, in John's footage sky is mostly in the same tonal range, there are no gradations from deep blue sky to lighter sky in lower parts. I think that my 12k would exhibit the same behaviour as his in this scenario (Aria), no banding.

You can find it on John's blog:
https://johnbrawley.wordpress.com/2022/ ... ter-tests/

Direkt link to the files download area John provided:
https://bit.ly/3TyIIIW

Thanks for sharing. I’ll take a look when I’m back home.

Hey guys!

Just tested it again in all compressions. From Q0 and 5:1 to lower. All of them exhibit this artefacts, no matter of compression or Constant Bitrate or Constant Quality...
Nice!

I've got a drastic further test suggestion, Mario, before you throw your lovely 12K in the rubbish; you may curse me and people jump on me if it doesn't work out I'd be inclined to try it - take that Rawlite out and test again.

Hey guys!

Just tested it again in all compressions. From Q0 and 5:1 to lower. All of them exhibit this artefacts, no matter of compression or Constant Bitrate or Constant Quality...
Nice!

Well, that is very disappointing to hear. I am out of ideas now and it really would be cool if someone from BMD could chime in here.

I've got a drastic further test suggestion, Mario, before you throw your lovely 12K in the rubbish; you may curse me and people jump on me if it doesn't work out I'd be inclined to try it - take that Rawlite out and test again.

Steve's suggestion may be worth the effort as I have an URSA Mini Pro 12K (no OLPF) and have never seen any such banding from it. Granted that I really only use Constant Quality. p24 Q0 at 12K when doing landscape shots and for wildlife typically 6K using p60 Q0 or 120fps Q1. I have used 12K p24 Q5 for a long interview and it worked well. I have done some tests with Constant Bitrate but prefer the Constant Quality.

Well, that is very disappointing to hear. I am out of ideas now and it really would be cool if someone from BMD could chime in here.

Hopefully someone will Robert but they haven't up till now and Mario has had no joy from support; despite JB's and your sage advice, it could be some time if ever. I'm no DOP but from my post experience I have encountered this before (see example which is moiré), and I am sure it's nothing to do with compression banding but some form of moiré; which would possibly alter with compression ratios. It could be that the Rawlite is not fully matched or fitted; in any case it's the only thing different about this camera from stock (i.e not BMDs own OLPF, as I understand it) and it would be worth it to me to eliminate it. It could be from a certain batch, who knows. Hans is very receptive to feedback and if taking it off does eliminate it, he will almost certainly change it.

moire_chris.jpg (108.67 KiB) Viewed 605 times

The other test would be a shot that changes the amount of sky in the shot.

100% sky, 90 % sky etc.

I do wonder if the codec is tripping on the amount of sky.

JB

The other test would be a shot that changes the amount of sky in the shot.

100% sky, 90 % sky etc.

I do wonder if the codec is tripping on the amount of sky.

JB

I don’t think it is related to the amount of sky, although it might provoke it sheer by the change you catch the sun as well.
The file ‘12KAriaDAY_21mm_BlackMist_1_8.braw’ shows banding as well when it’s exposed to extremes. The sun is just behind the cactus. See picture.

12KAria-banding.jpg (304.61 KiB) Viewed 539 times

The other test would be a shot that changes the amount of sky in the shot.

100% sky, 90 % sky etc.

I do wonder if the codec is tripping on the amount of sky.

JB

I don’t think it is related to the amount of sky, although it might provoke it sheer by the change you catch the sun as well.
The file ‘12KAriaDAY_21mm_BlackMist_1_8.braw’ shows banding as well when it’s exposed to extremes. The sun is just behind the cactus. See picture.

12KAria-banding.jpg

Hang on. Thats a silly grade to say hey there’s banding !!

I mean I’m not looking at the OPs files but no file is going to survive that kind of treatment.

JB