New Ursa 12k sensor tech

The place for questions about shooting with Blackmagic Cameras.
  • Author
  • Message
Online
User avatar

Dmitry Shijan

  • Posts: 1644
  • Joined: Wed Sep 17, 2014 7:15 pm
  • Location: UA

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 12:54 pm

1. I guess if you shoot 12K you will see some moire because it is normal for 1:1 pixel capture for any camera.

2.If you shoot 8K downsampled in-camera and put it to 4K timeline you probably also see a moire because some specific mismatch between BRAW processing and timeline resolution.

3. So as it was explained at slashcam.de article, with Ursa 12K if you shoot 8K, you need to put it to 8K timeline. Or if you shoot 4K you need to put it to 4K timeline. This should not produce moire.
All my custom made accessories for BMMCC/BMMSC now available here https://lavky.com/radioproektor/
Offline

Andrew Kolakowski

  • Posts: 6842
  • Joined: Tue Sep 11, 2012 10:20 am
  • Location: Poland

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 2:52 pm

This should not be the case- this is Resolve engine issues and should be fixed.
Working on 8K source with 4K timeline is very different than on 8K in 8K timeline (and scaling on export). 2nd is way more resource demanding. This would be an annoying limitation.
Offline

John Brawley

  • Posts: 2674
  • Joined: Tue Aug 21, 2012 7:57 am
  • Location: Los Angeles California

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 4:06 pm

Resolve seems to work better than Premier. Isn’t that what Slashcam are saying ?

JB
John Brawley ACS
Cinematographer
Currently - Los Angeles
Offline

John Griffin

  • Posts: 851
  • Joined: Sun Aug 23, 2015 4:18 pm

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 4:14 pm

I still don't think the moiré I saw is anything to do with how the Resolve engine (or Premier) is handling the footage and is how the camera is processing it. BRAW is not RAW data from the sensor and is partially processed in camera.
i7 6950x, 64gb RAM, GTX1080ti.
Offline

John Brawley

  • Posts: 2674
  • Joined: Tue Aug 21, 2012 7:57 am
  • Location: Los Angeles California

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 4:23 pm

John Griffin wrote:I still don't think the moiré I saw is anything to do with how the Resolve engine (or Premier) is handling the footage and is how the camera is processing it. BRAW is not RAW data from the sensor and is partially processed in camera.


I agree that the moire that I’ve seen in my own work is intrinsic to the 12k, and like any camera that doesn’t have an OLPF.

No way is it as bad as what’s out there as examples that we’ve all seen. My view aligns with what Slashcam are saying. It’s negligible.

That’s all I’m trying to reconcile.

JB
John Brawley ACS
Cinematographer
Currently - Los Angeles
Offline

John Griffin

  • Posts: 851
  • Joined: Sun Aug 23, 2015 4:18 pm

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 5:04 pm

Yes it's negligible but is it actually classic moiré? ( pixel peep warning)

Password - 'moire?'
Changing decode quality in Camera Raw setting does nothing so set it at same as project. 5x zoom in a UHD timeline no grading other than ACES REC709 and exported with stock vimeo UHD settings from Resolve. It's an 'artefact' for sure but the rest of the fine details and edges in the image are clean where I would expect to see aliasing as well.
Again - this BRAW footage I've downloaded and played with in Resolve is amazing for the shear artefact free clean detail and this is just nerdy stuff pulling this up really and yes you will need top notch glass to get this level of detail but maybe less than top notch glass will hide this artefact? Better not look at it much more as my P6k is arriving tomorrow and it may be a let down......
i7 6950x, 64gb RAM, GTX1080ti.
Offline

Andrew Kolakowski

  • Posts: 6842
  • Joined: Tue Sep 11, 2012 10:20 am
  • Location: Poland

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 5:26 pm

John Griffin wrote:I still don't think the moiré I saw is anything to do with how the Resolve engine (or Premier) is handling the footage and is how the camera is processing it. BRAW is not RAW data from the sensor and is partially processed in camera.


1 RAW recording proves nothing at all. Moire is complex subject. Any camera without OLPF will have moire to some degree. Moire introduced due to "bad" processing is another story.

RAW from 12K camera is very different to all others, which use Bayer pattern. Ideally it needs new debayering technique to get best out of new pixels layout. Regardless of this there can be other issues, eg. in Premiere (or in Resolve as well) etc. which caused all this 'so bad' moire.
Last edited by Andrew Kolakowski on Wed Nov 18, 2020 5:32 pm, edited 1 time in total.
Offline

John Griffin

  • Posts: 851
  • Joined: Sun Aug 23, 2015 4:18 pm

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 5:32 pm

Andrew Kolakowski wrote:
John Griffin wrote:I still don't think the moiré I saw is anything to do with how the Resolve engine (or Premier) is handling the footage and is how the camera is processing it. BRAW is not RAW data from the sensor and is partially processed in camera.


1 RAW recording proves nothing at all. Moire is complex subject. Any camera without OLPF will have moire to some degree. Moire introduced due to "bad" processing is another story.

RAW from 12K camera is very different to all others, which use Bayer pattern. Ideally it needs new debayering technique to get best out of new pixels layout.

Which 'bad processing' do you mean exactly?
i7 6950x, 64gb RAM, GTX1080ti.
Offline

Andrew Kolakowski

  • Posts: 6842
  • Joined: Tue Sep 11, 2012 10:20 am
  • Location: Poland

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 5:34 pm

For example nearest neighbour scaling behind your back. We have no idea what Premiere's engine is doing when timeline is 4K and source footage 12K.
Any conclusion about camera native moire should be held until some assurance is given that handling 12K RAW in Resolve/Premire is not causing native moire to be 'boosted'. This can be very possibly the case.
Offline

John Griffin

  • Posts: 851
  • Joined: Sun Aug 23, 2015 4:18 pm

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 6:00 pm

Andrew Kolakowski wrote:For example nearest neighbour scaling behind your back. We have no idea what Premiere's engine is doing when timeline is 4K and source footage 12K.
Any conclusion about camera native moire should be held until some assurance is given that handling 12K RAW in Resolve/Premire is not causing native moire to be 'boosted'. This can be very possibly the case.

If you set the raw input to 12k and then zoom it’s not downsampled in my understanding but I could be wrong. Did that and used every type of resize option ( there are lots now in R17) I assume you have downloaded and played with the BRAW clip and did you see any changes?
i7 6950x, 64gb RAM, GTX1080ti.
Offline

Howard Roll

  • Posts: 967
  • Joined: Fri Jun 03, 2016 7:50 am

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 6:20 pm

Every pixel in the 12K sensor is derived from 4 filter kernels, a hi, lo, luma, and chroma. Each 2D Gaussian filter samples neighboring pixels differently prior to the 4:2:0/18:1 Braw encoding. My guess is that both softness and moire are a function of this interaction as the image is assembled from the discrete maps.

The Dumbo shot is a treat, worth the wait. Here are some real observations looking at the effects of timeline and decode scaling. I wonder how 4K on the wire compares?

Good Luck

14K.jpg
14K.jpg (242.12 KiB) Viewed 498 times
Offline

John Griffin

  • Posts: 851
  • Joined: Sun Aug 23, 2015 4:18 pm

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 6:38 pm

12k.jpg
12k.jpg (270.49 KiB) Viewed 482 times
With a timeline setting of 12288x6480 (and my GPU complaining) the artefact is still there. Screen grab from UI.
i7 6950x, 64gb RAM, GTX1080ti.
Online
User avatar

rick.lang

  • Posts: 13981
  • Joined: Wed Aug 22, 2012 5:41 pm
  • Location: Victoria BC Canada

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 7:11 pm

Howard Roll wrote:Every pixel in the 12K sensor is derived from 4 filter kernels, a hi, lo, luma, and chroma. Each 2D Gaussian filter samples neighboring pixels differently prior to the 4:2:0/18:1 Braw encoding...

14K.jpg


When I look at the rendering of text, the 12K debayer and 4K timeline looks best to me. The 4K debayer and 4K timeline does not look as good as I assumed based on CaptainHook’s guidance to use the Camera raw debayer setting. It would seem setting the timeline to 4K does something differently than setting the Camera raw debayer to 4K.

Is that what you see Howard?
Rick Lang
Offline

Andrew Kolakowski

  • Posts: 6842
  • Joined: Tue Sep 11, 2012 10:20 am
  • Location: Poland

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 7:40 pm

4K timeline will probably take only 4K from 12K at RAW level (to save resources) and this is why it looks much softer.
When you specifically set 4K debayer (can you do this?) out of 12K then this is not good choice I think. Probably worse than recording 4K natively.
Debayering 12K to get 4K final image should always be the best choice, but it will hit performance.
Last edited by Andrew Kolakowski on Wed Nov 18, 2020 7:53 pm, edited 1 time in total.
Online
User avatar

rick.lang

  • Posts: 13981
  • Joined: Wed Aug 22, 2012 5:41 pm
  • Location: Victoria BC Canada

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 7:53 pm

Andrew Kolakowski wrote:...
Debayering 12K to get 4K final image should always be the best choice, but it will hit performance.


Thanks for confirming what I saw. I was hoping that wasn’t true; what Howard posted means there’s not likely anything to be gained from posting a video of the testing I planned yesterday. I’ll just play with that 12K footage for my edification looking at the results on my 6K screen. Perhaps I’ll notice something worth reporting.

I don’t have the UMP12K so it’s an existential angst I’m feeling as it appears shooters really should record 12K in camera when they require the best possible 4K deliverables. There’s no free lunch for the most demanding work.
Rick Lang
Offline

Howard Roll

  • Posts: 967
  • Joined: Fri Jun 03, 2016 7:50 am

Re: New Ursa 12k sensor tech

PostWed Nov 18, 2020 9:49 pm

rick.lang wrote:When I look at the rendering of text, the 12K debayer and 4K timeline looks best to me. The 4K debayer and 4K timeline does not look as good as I assumed based on CaptainHook’s guidance to use the Camera raw debayer setting. It would seem setting the timeline to 4K does something differently than setting the Camera raw debayer to 4K.

Is that what you see Howard?


I prefer the 12K decode in the 4K timeline as well. The 12K timeline has a little more detail/contrast also more noise.

From slashCAM
It was also interesting that the 12K recordings decoded in 4K could not be distinguished from "real" 4K recordings from the camera. Resolve obviously uses the same algorithm in the camera and in the software.

I think Hook also said that the 4K decode is only sampling 1/9th the pixels. Without the benefit of upscaling, the results seem expected. The 12K decoded footage is zoomed out .3X in a 4K timeline which smooths some of the noise. Shooting for 8K delivering at 4K looks to be the happy spot.

Good Luck
Offline

newbie

  • Posts: 14
  • Joined: Fri Nov 06, 2020 5:23 am
  • Real Name: Dan Lane

Re: New Ursa 12k sensor tech

PostSat Nov 21, 2020 6:18 am

First thing I did with my 12k was shoot a bunch of charts to fine tune my exposure and select my lens set in preparation for client work. I did find that the 12k was nearly maxing out my 52" wide ISO 12233 resolution chart.

Note - I don't have a lab, these test charts are only for my personal use and don't represent the sensor or optics.

Screen Shot 2020-11-20 at 10.13.22 PM.png
Screen Shot 2020-11-20 at 10.13.22 PM.png (925.68 KiB) Viewed 219 times


But I don't see a huge difference between shooting 12k and 8k as far as the recording source. I took 12k renders in to photoshop and downscaled them to 8k using photoshop to oversample and again, I didn't see significant difference between 12k to 8k and recording to an 8k source file.

So that tells me the built in algorithms are pretty good. I still shoot 12k files since I've been able to do some extra tricks in preprocessing.

Is a Ursa 12k oversampled 8k better than a red monstro at 8k? I'd be interested to learn that. I'd like to get them side-by-side looking at the same charts, under the same light, and using the same glass.

A side note.. I have not (yet) personally had any moire issues on shoots.
Attachments
Screen Shot 2020-11-20 at 10.10.16 PM.png
Screen Shot 2020-11-20 at 10.10.16 PM.png (980.07 KiB) Viewed 219 times
Offline

John Brawley

  • Posts: 2674
  • Joined: Tue Aug 21, 2012 7:57 am
  • Location: Los Angeles California

Re: New Ursa 12k sensor tech

PostSat Nov 21, 2020 6:33 am

Dan these look like screenshots, are they from final renders or from within the viewer in Resolve ?

JB
John Brawley ACS
Cinematographer
Currently - Los Angeles
Offline

newbie

  • Posts: 14
  • Joined: Fri Nov 06, 2020 5:23 am
  • Real Name: Dan Lane

Re: New Ursa 12k sensor tech

PostSat Nov 21, 2020 7:11 am

They're screenshots of tiff renders. I've got a 1mb upload maximum here so I'm having to crop the image significantly to keep it 1:1.

Just looked back at the render folder and the next few frames actually appear slightly sharper.

Screen Shot 2020-11-20 at 11.03.52 PM.png
Screen Shot 2020-11-20 at 11.03.52 PM.png (543.22 KiB) Viewed 189 times
Attachments
Screen Shot 2020-11-20 at 11.03.52 PM.png
Screen Shot 2020-11-20 at 11.03.52 PM.png (543.22 KiB) Viewed 189 times
Previous

Return to Cinematography

Who is online

Users browsing this forum: Csaba Nagy, Google Feedfetcher and 23 guests