Black noise artefacts problem with Resolve (and not ACR)

[Rojiraan]

Hi all !

So I just started making some RAW video tests with my 5DMk2 using the latest build available (Nightly.2016Apr01). I shot various things, and while trying to develop a concrete workflow for a future project, I discovered something strange.

Here is a video explaining everything : https://youtu.be/tGiljKKBQjs

Basically, I have some black noises while grading in Resolve (the problem didn't appear in After Effect, the noise looks way better inside AE). Apparently, the problem is only concerning two parameters : Color Temp and Tint. The others parameters don't seems to change anything to the problem.
The strange thing I didn't show in the video is that if I reduce the decode quality of the footage, the noise disappear immediately. I should always see some of these black pixels even at half quality, no ?
I searched a little bit on the forum, but I didn't saw anything speaking of this.
Also, the problem is directly visible using the MLRawViewer software. Opening the DNG conversion from the MLV inside any Photoshop or After Effect (so using ACR) remove the problem.

And, that's it. Have you ever been facing this problem ?
Thanks !

[cmccullum]

From what I understand, Adobe handles debaying better than resolve, this may be the reason these artifacts become visible.
As far as lowering the quality the artifacts are probably just getting "covered"
In mlrawviewer, try changing the debayer algorithm see what effect that has
Lastly, in your video, there are still artifacts in the same place in the AE example, but they don't have the colour

[Rojiraan]

Hey ! Thanks cmccullum for this fast answer !

I tried various things in MLRawViewer, but nothing is changing the noise quality (even when exporting a ProRes file).
Same thing using MLVFS, I tried to enable some "hacks" to see if something was evolving inside Resolve, but nothing.

Yep I agree with you speaking of the noise artifacts, but... I don't know, the noise seems to be "processed" in a way more natural way inside ACR, I think. I don't know why these blacks artifacts are so visible inside Resolve 

[cmccullum]

Not a problem. I'm definitely not a pro when it comes to this stuff, but I'd say I've got reasonably decent problem solving skills so I try to help out whenever I can :p

When you say you "tried various things" in mlrawviewer, what do you mean exactly? Did you change the debayer type?

Also, in resolve, do you have highlight recovery on?

[Danne]

That is some strange looking dots. Maybe something that darkframe averaging could fix?
http://www.magiclantern.fm/forum/index.php?topic=13512.msg164871#msg164871

Is the dot still visible after conversion in DaVinci Resolve? If you could upload a sample file we could have a look at this.

[Andy600]

A sample DNG would be useful but a couple of things I've noticed from the video:

1. A nearly 4 stop increase in exposure is quite a lot. The shot is likely under exposed by around 3 stops so you have pushed a lot of the mids into the noise floor and the artifacts could be caused by shadow pixels being shifted to zero.

2. What app did you use to create the DNG. MLRawViewer produces lossy DNGs (I think). Use another converter.

3. White balance requires some big shifts which points to the color matrices being either only a single daylight illuminant matrix (i.e. DCRaw Adobe coefficient) or possibly wrong color matrix tags altogether. If either of these is the case then you cant use the same white balance settings in Resolve and ACR. In some raw2dng apps a custom white balance setting will not be saved so check what the meta is telling you. Sometimes it's better to just select an in-camera preset white balance i.e. the one that is closest to the lighting you are shooting under.

4. You are white balancing by eye without any reference. Egg white under tungsten illumination (or any other light for that matter) is not pure white. Get a color chart or at least a good photographic white card AND definitely get a gray card for getting your exposure correct. Then if you choose to ETTR or under-expose the shot you will at least know by how much and can pull it back properly in post.

5. Unfortunately, I think this one is probably a lost cause because of all the above and I doubt dark frame subtraction or noise reduction will help much.

[Rojiraan]

Sorry for the delay.

I uploaded 3 new MLV files with basically the same lighting configuration. Each different files represent another ISO (400,800 and 1600). All the videos have been shot @ 4000K.
Here are the files : https://drive.google.com/file/d/0B8OXHTxcIY-6WEF0SXR2NVVscGc/view?usp=sharing
On Resolve, playing with the Tint/Color Temp parameters shows that the black noise artifacts are really visible on the 400 ISO file. But even at 1600 ISO, I feel that the noise quality is not really beautiful on Resolve/MLRV.
Well, finally, I just think that ACR noise processing looks way more beautiful than Resolve (at least, for me). But, is it the same with, for example, R3D files ? This is starting to scares me a little bit... I need to try. I also saw on the post processing subforum a page speaking about a software called Pomfort Cliphouse, I will download it when I have the time it to see how the MLRaw is processed with this.

@cmccullum -> Yep, I changed the debayer type, changed also the stripe correction, in Resolve I tried pretty much every Camera RAW tools (including Highlight Recovery). It seems that in this last software, only the Tint and the Color Temp parameters are makings these dots appear/disappear. By the way, I just noticed that playing the file with Bilinear debayering (so inside MLRV) seems to create a softer image (less sharpness, less noise visible). If I pause the file, the black noise is back.

@Danne -> Thanks for the link, didn't know about this darkframe issue. I will take a look into this. And yes, even after a ProRes/DNxHR render with Resolve, the dots are always visible.

@Andy600 -> Thanks a lot for these indications. In fact, I was just trying the 5D RAW capabilities without doing any concrete tests, and that's when I decided to examine a possible post-prod workflow using these footage that I noticed this noise quality difference between ACR and Resolve/MLRawViewer.
About the color temperature settings, I was thinking that it was possible to deal with this in post (as always with RAW files if I'm correct, sure it's better when you 'white balance' your shot correctly I suppose ), so I didn't bother too much with this (the egg you saw on Youtube have been shot using Auto WB).
I don't really understand when you're speaking about the color matrix. If I'm correct, you mean that the color matrices of these softwares aren't up-to-date (or that they don't correspond with the 5D RAW color matrix) is that right ? I need to do further documentation about this. By the way I effectively saw that, for example, a +150 tint in ACR are not doing the same thing than in Resolve, as shown in the Youtube video.
Yes, I used MLRawViewer to produce the DNGs, but as said before MLVFS did the same thing (by the way, I don't understand why but I can't see the virtual drive appear launching MLVFS.exe, even after upgrading to Dokan 1.0.0RC2 and using MLVFS x64, but that's another problem ).

Thanks again for the help and documentations !