Blackmagic Forum

I’m sure we’ve all been here before, at least I’m hoping that’s the case…

So we wrapped our short film, shot on the (OG) Blackmagic Micro Cinema Camera, and honestly the whole production was a dream come true. We watched dailies every night and everything seemed fine, until it came time to look at our interiors. Unfortunately, by the time I looked at these scenes closely, re-shoots were no longer possible.

For our interior scenes, I made the mistake of both shooting at ISO 800, and not ETTR to get as much light onto the sensor as possible. This resulted in some pretty nasty chroma noise, luminance noise (which doesn’t bother me too much as we want an s16 look anyway), and Fixed Pattern Noise.

I haven’t applied any color grading yet, just basic WB correction (we stayed at 4500k) but the noise is freaking me out so much I’m afraid it’ll be hard to edit the footage.

I’ve been trying to learn as much as I can about Noise Reduction, and feel as though I’m starting to get a pretty good handle on it. I know that ultimately the move will be just to lean in to and embrace what we have.

This whole thing has been really hard on me, and I can definitely say it is lesson learned, but I figured I would come here for perhaps more specific tips or tricks. Any and all help is greatly appreciated as this is essentially the “Senior Thesis” of our self-made film school.

Sorry for the long post, and thank you so much in advanced for potentially saving our project and my sanity.

In which format did you record?

You can upload a few of your problematic shots if you want here and I'll have a look about what to do:
https://laufbildkommission.wetransfer.com

We recorded in Raw 3:1 cDNG

Thank you so much!!

One tip, after you denoise the heck out of it, add some grain back in. It will hide the smoothing of the noise reduction and make it look like there's detail there. Goodluck!

Neat Video has a bit (not too much) of a learning curve, but is incredible. And at only $99 I think is still the industry standard. (You'll see Neat Video watermarks on HBO.)

https://www.reddit.com/r/vfx/comments/d ... stream_of/

You can choose what percentage to leave on. For difficult shots, I leave it on at 75% and it feels like I gained two stops of sensitivity/cleanness.

You should be fine.

How do you avoid shots looking smudgy or plasticky with Neat Video?

I’m sure that’s part of the learning curve you’re talking about

Watch the tutorials. Use more temporal denoising than spatial. Use it at 75% strength instead of 100%. (Or re-grain after.)

Neat Video

Neatvideo here too.
Additionally, you can reduce FPN by shooting a reference with the same settings and even similar temperature, and the lens covered. The result can then be subtracted from your footage.
The strongest point of Neatvideo is temporal NR, which doesn’t help with FPN.

NeatVideo has been such a life saver, thank you guys so much for the suggestion!

One question for you all is- when applying NV, do you raise the exposure, apply NV to that and then bring it back down to the original exposure, OR just apply NV to the image as is without touching exposure

Thanks again!

I mostly use this for denoising:
https://github.com/WolframRhodium/VapourSynth-BM3DCUDA

However, it is very compute intensive and requires Vapoursynth which does have a learning curve.

For the most difficult noise, Neat video works best in linear.

I apply a Color Space Transform from my camera log space to linear, then apply Neat Video set to its linear mode, then another Color Space Transform from linear back to my camera log space.

You should see better results working that way than applying Neat Video’s “gamma” mode to footage in log.

Thanks Jaime!

I actually JUST seeing that I missed this step…

In using the BMMCC (OG), is there anyway you could give me an ELI5 (or I guess step-by-step process) of how what you just described works? A lot to ask, I know, so I understand if you don’t feel like doing that haha

Thanks in advanced!

In the raw tab of Resolve — to what color space and gamma are you decoding the cDNG files from your BMMCC?

Jaime, can I embarrassingly expose myself and say that I don’t really know…

Edit to clarify: I know roughly what you’re referring to, but am unclear as to what it should be

Hey Jamie,

So right now I have

Input Color Space - Blackmagic Design Film Gen 1
Input Gamma - Blackmagic Design Film
Output Gamma Linear

But I was unsure what my Output Color Space should be? Is it Rec. 709?

The node tree order would be:

Color Space Transform
Input Color Space — Blackmagic Design Film Gen 1
Input Gamma — Blackmagic Design Film
Output Gamma — Linear
Output Color Space — Blackmagic Design Film Gen 1
(all other options disabled or set to none)

to

Neat Video (with the "input data type" option set to "Linear")


to

Color Space Transform
Input Color Space — Blackmagic Design Film Gen 1
Input Gamma — Linear
Output Color Space — Blackmagic Design Film Gen 1
Output Gamma — Blackmagic Design Film
(all other options disabled or set to none)

Jamie thank you so much I really appreciate that!

One last question is, what should my settings in DaVinci be for the color space of the overall project?

Texaco87 wrote:Jamie thank you so much I really appreciate that!

One last question is, what should my settings in DaVinci be for the color space of the overall project?

You can leave it at default and do your colorspace settings at node level.
As with everyone, use NeatVideo - it will help a lot, plus the node tree order Jamie gave.
Adding some grain in the end will help hide the plasticky look. If that's not enough, a very slight mist or vapor overlay can help masked but use it very, very sparingly. This idea is using creative styling to your advantage to mask problems like this.

Ellory, thanks for the reply

I'm afraid I've changed the settings from default at this point, is there somewhere I can find what they should be for the BMPCC OG / BMMCC?

The unmanaged default vs Resolve Color Management is well covered in the Resolve manual:
https://documents.blackmagicdesign.com/UserManuals/DaVinci_Resolve_18_Reference_Manual.pdf?_v=1702368010000

And in Blackmagic Design's free training videos:
https://www.blackmagicdesign.com/products/davinciresolve/training

The Color Space Transform settings in my previous post were for the unmanaged default project setting.
If you instead have your project set to Resolve Color Management, it would be:

Color Space Transform
Input Color Space — Use Timeline
Input Gamma — Use Timeline
Output Gamma — Linear
Output Color Space — Use Timeline
(all other options disabled or set to none)

to

Neat Video (with the "input data type" option set to "Linear")

to

Color Space Transform
Input Color Space — Use Timeline
Input Gamma — Linear
Output Color Space — Use Timeline
Output Gamma — Use Timeline
(all other options disabled or set to none)

Seems like I have some reading to do, thanks again Jamie

Uli Plank wrote:Neatvideo here too.
Additionally, you can reduce FPN by shooting a reference with the same settings and even similar temperature, and the lens covered. The result can then be subtracted from your footage.
The strongest point of Neatvideo is temporal NR, which doesn’t help with FPN.

Uli, are you adding the resulting FPN reference frame as a matte to the color page node tree and doing the subtraction within a pre-clip group grade?

Seems like an even better idea, Jamie, but I use it simply on its own track over the footage to treat.
To generate the FPN picture, I record covered for about 30 seconds and then time-average the result to get rid of dynamic noise.

Thank you for the details Uli. What method are you using to time average the reference frames?

Probably not very helpful if you don't use their plug-ins anyway, since it would be costly to get for only this purpose: S_TimeAverage by BorisFX.
I think you should research in forums for astrophotography, I'm sure there are other options.

Ah, so there is a secret to the sauce. I appreciate the details and will look to what alternatives exist. I wonder if there is a way to yield a similar time average using the tools in Fusion.

Manual averaging works like this:

First layer transparency 100%
Stack a second layer with 1/2 = 50%
Third layer with 1/3 = 33,33%
Fourth layer with 1/4 = 25%
Fifth layer with 1/5 = 20%
Sixth layer with 1/6 = 16.67%
Seventh layer with 1/7 = 14.29%
Eight layer with 1/8 = 12.5%
Ninth layer with 1/9 = 11.11%
Tenth layer with 1/10 = 10%

and so on and so forth

Each layer has been shifted one frame in time to one before of course.

You don't need NEAT Video, you can use Resolve's updated noise reduction.

Dermot Shane (I think he's also here) posted comparison results against Neat Video and DVO_Velvet over at LiftGammaGain and quality is on par but faster.

https://www.liftgammagain.com/forum/ind ... 640/page-2

Also, I once directed a commercial shot on OG BMPCC and the DP accidentally underexposed everything by 1.5 stops. But It still looked great after grading - we didn't even apply noise reduction. All hope is not lost

Michel Rabe wrote:You don't need NEAT Video, you can use Resolve's updated noise reduction.

Dermot Shane (I think he's also here) posted comparison results against Neat Video and DVO_Velvet over at LiftGammaGain and quality is on par but faster.

Sorry, have just read that thread through and as someone who has specialized in denoising and all sorts of image restoration with experience of over 20 years now, Neat Video is still the king in town. If Dermot Shane needs 10-12 minutes to get the settings right, then I have my doubts about his competence with Neat Video.

Dermot Shane wrote:

- Neat took about 2 min to open UI, analise, tweak settings, close UI, cache at 8fps, chk on a real monitor - then rinse and repete twice until in the sweet spot, around 10-12 min of suite time, the lack of interativity really cost time, end resualt was in the zone, orignal analysis was not all that great, smearry artifacts on the umbrella against building in the bg + plastic skintones, got that sorted, but at the cost of a lot of suite time

Nothing against Dermot - but this quote tells me he is not that fit in NV.

For almost all use cases it makes more sense to first try Resolve's NR before buying NEAT.

From my experience you can get pretty far with DR’s NR if it’s mild noise.
For tough cases, it’s still NV for me.

Uli Plank wrote:From my experience you can get pretty far with DR’s NR if it’s mild noise.
For tough cases, it’s still NV for me.

+1. I have both. I use DR but with really bad noise especially with the chroma (subjective), in my experience, pushing temporal and spatial in DR makes the image look plasticky. With Neat, there’s more dialing in that provides a more natural look with the fix.

Ultra NR in Resolve 19 is a new one though.
Seems to be good already and will only improve in the future.

I use Neat Video for 16y and it likely still has it's place if you already own it, but for the vast majority of times Ultra NR will probably be good enough before buying Neat.

Ultra NR needs to get a lot faster to compete with NeatVideo. About 5 to 10 times!

Guys, I can't tell you how much I appreciate all of your help and guidance. Neat Video truly is a life saver.

One question I have for you all is - is there any reason not to just set Temporal Radius to 5 frames? I don't mind the render time, but it seems like this results in the cleanest footage possible with the least amount of loss of detail.

Follow up to this - when I did this (and set both quality options to high), the clips still playback choppy even after cacheing the clips. Is this just because so much processing is being asked of the program/computer? Also, if I leave it like this, will the clips still be choppy after rendering the entire timeline out?

Hope I phrased this right. Thanks again!

I find that generally unnecessary, 2 or 3 works quite well for me. I have a hard time to see a difference between 3 and 5, but the rendering times get really long. OTOH, I suppose it all depends on the structure and severity of the noise, so just trust your eyes. You know that under Advanced you can study pretty well what NeatVideo does, do you?
Regarding caching, I have no explanation. Other than, maybe, your codec setting is too massive and your storage can't keep up? How does it look like when you Render in Place? Anyway, final renders should be smooth. Last week I had a very long project with a lot of NeatVideo, there was no issue in the final result at all.

Hey Uli,

Thanks for the quick response buddy

I do know about Advanced mode...it took me about 2-3 weeks of working with/studying NV on & off to learn that the best application is the simplest/most straightforward one in terms of how to use it

I will go back and see how it looks with 3 vs. 5 and maybe readjust

But I think you're right - I have 1080p raw 3:1 cDNG footage that is being processed, so I'm sure that's it

Thanks for all that practical real life experience and advice!

I’m storing a copy of the thread locally on my server, if That’s OK!

Regards

Texaco87 wrote:One question I have for you all is - is there any reason not to just set Temporal Radius to 5 frames? I don't mind the render time, but it seems like this results in the cleanest footage possible with the least amount of loss of detail.

R = 5 delivers the cleanest footage especially with lower resolution like HD but be careful - the way it works it can change an actor's lip movement slightly for example.
But it will carve out a lot of details in dark hair and other structures hidden in the noise.

Texaco87 wrote:Follow up to this - when I did this (and set both quality options to high), the clips still playback choppy even after cacheing the clips. Is this just because so much processing is being asked of the program/computer? Also, if I leave it like this, will the clips still be choppy after rendering the entire timeline out?

Per default caching is set to DNxHR HQX which delivers superb quality but is demanding in file size.
I found playback will improved with lighter codec versions like DNxHR SQ in your render cache settings.

Robert (and Uli and others), do you have a preference for putting noise reduction as the first node or the last node in the tree? Some people put it somewhere in the middle, but I usually use it as the first node so it's working on the original signal, although I've read that NR can affect later qualifications so some people advocate putting it at the end.

I've seen logical arguments for either approach; I think the main drawback to putting it as the last node is that it has to recache if you make changes to earlier nodes. I sometimes just disable the NR node altogether for faster playback, rather than caching, and re-enable it for the final render.

I normally put NR as the first node, but it also depends on your color science. I found with DVR internal NR it works well with Color Science 4 because that has a bit more contrast and with Color Science 5 it works better later in the chain. With UMP12k footage just a little Chroma NR works very well in the first node. So it really depends.
If the image is too flat, the algorithm has troubles to determine noise.

NeatVideo is different and I found no differences where I place it, so usually I keep it in the first node.

Try Resolve's UltraNR.

Curious too. It’s dead slow on Apple silicon, so I’d like to know how it compares under Windows or Linux.