Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Where should Noise Reduction come in the post process?

Tim Hole

Tim Hole

Well-known member
Joined
Jun 24, 2008
Messages
1,327
Reaction score
0
Points
0
Age
48
Location
United Kingdom
I am getting more than one person advising me on changing my process on Noise Reduction so I am asking those more in the know...

I have always thought it logical to get the footage to its best point before the grade, however I have people telling me otherwise. So is it best to do it before grade, after primary grade or after the secondary. Does it even matter? Does the bit-depth have any influence in this decision? Does the process change depending on which program you use ie. Neat Video, FilmMaster, MB, Topaz etc...

If you were denoising 8-bit would you denoise at a higher bit depth? Would you benefit from this?

All these seem rather unclear to me.

Thanks in advance for your thoughts.
 
I would like to revisit this topic.

My question is which version to purchase in regards to Neat Video.

It seems that for me, I would have to buy both After Effects version and NLE version in order to work across all workflows.

Of course the topic in question is most important to me as well.

So when do we apply the noise reduction?
 
I asked since in some cases I cant do post work on noisy clips.

If Im doing compositing in After Effects with Maya datasets its really hard to work with when the plate is buggered up.
 
typically noise reduction/sharpening is the final-final procedure...
that being said, some people do it first thing and are happy
altering the entire image's texture (nr/sharpen/compress) can hide grading/fx work
if you're working on your own you can do whatever you want, but i would always consider your final output(s)
 
David Mullen ASC said:
Since the grade may add more noise to a shot, I would think after the grade.
Click to expand...
That makes sense to me, too. The problem as I see it is if the grade really stretches out the exposure quite a bit, it's going to exaggerate grain on those specific shots.

Note that the real point of grain reduction is not a "one size fits all" setting, but rather, a shot-by-shot adjustment that makes the apparent grain consistent throughout the entire scene. This is why a lot of facilities refer to this as grain management rather than grain reduction​.

I've also found that it sometimes makes more sense to do more grain reduction in certain colors than others -- for example, if you have a lot of blue noise in the blacks, then crank up blue-only noise reduction there, and leave the other color channels alone. (Essentially pull a key and do NR on the key.) In the case of noisy blues in the sky, it's not that hard to defocus just the blue channel a tiny bit, and that takes the edge off the grain without using NR per se.
 
There is one benefit with using for example Neat Video before the grade. Neat Video rebuilds the image in the frequency spectrum and can actually take way banding and create smooth gradients. I used to use it a lot on Canon 5D footage to reduce compression artifacts and nasty banding. When you do this, its very important to increase the bit depth to 16bit before. Basically this would transform your 422 8 bit compressed image into something reminding of a 444 16 (ish) bit depth image.

I haven't used Neat Video on Epic shots yet, but I have used Neat Image to take away banding on 16 bit stills images and I would think the quality improvement should be similar there.

I also found it to be very useful to view the Luminance in the frequency spectrum in Neat Video to really see what was going on in the noisy areas. Myself I would also use Neat Video/Image to sharpen images, often in combination of other sharpening programs.

When it comes to these workflows I have found that there is only really one good way of finding out, testing it yourself. To summarize I would first test noise reducing first. Maybe do the final sharpening pass last.
/Andreas
 
Depends on steps one's post workflow consists of.

For example, compositors may use noise reduction without your knowledge before keying or adding motion graphics, in which case you may get inconsistent texture between shots in final edit. In these and similar cases, where texture consistency can be affected by different people and processes, it may be helpful to have this factor under control and applied after first light. In most cases this process is best left as last one in the chain.

In addition, noise reduction can sometimes be avoided by good grading.
Sharpening is mostly a legacy of an HD acquisition era and I don't suggest using that route, unless a specific look intent is motivating it.
 
Over on the Avid User Forum, this question was asked of Neat Video and they said they recommend applying noise reduction before other image processing. They also said that grading doesn't strongly affect the noise signature, so applying NR after color grading shouldn't be a problem.

http://community.avid.com/forums/t/121955.aspx?PageIndex=2
 
To avoid oversimplification or misinterpretation of the linked post:

We usually recommend to apply noise reduction before any other post-processing steps to avoid aggravation of noise by those steps.
Click to expand...

If material has not passed primary corrections, noise is less visible and settings applied based on that visual inspection are practically guesswork and may not be sufficient for the end result. The other approach, with enhanced noise reduction and "playing safe" settings may negatively affect the appearance of texture, and may be unecessary if type of grading does not make the noise problematic.

Also, noise reduction applied in advance requires stepping out of raw and sending RGB to a grading suite and second processing of material prior to grading for no particular reason. And grading depends on material's "thickness" and precision, all of which get reduced with every processing. Nothing worth dramatizing over, with 12bit Prores and up, but still worth mentioning. There is also the data multiplication factor and maintenance after the project is done.

Some post-processing can make noise stronger or modify it, thus making it more difficult to reduce. Color grading is not one of those, so in principle it is okay to reduce noise after it too, but for simplicity sake it may be easier to apply noise reduction first in all cases.
Click to expand...

Very generally put, grading can enhance the visibility of noise. This is most common with enhancing global or local contrast, done improperly or even properly, but it can get more visible in other scenarios as well.
If processes such as sharpening, blurring or grain/texture management are used in a grading suite, noise is also being modified. It may also be slightly reduced by certain routes and in those cases it may require just a slight noise reduction applied at the end, on something which in the start appeared as needing more drastic NR.

Another thing worth mentioning is that NR applied prior to grading can affect grading decisions.
Food for thought.

Choices based on workflow simplicity and practicality greatly depend on the type of project. Any latter changes of heart, variations or alterations of the edit in case of pre-grade noise reduction require a greater fuss and additional rendering. In those types of scenarios noise reduction applied last in the chain removes a bunch of variables and reduces the time of the whole process.

There are far too many variables in this case to be able to suggest just one approach as the best one.
 
To me the only right answer to this question is to work node based from r3d source file to final image trough all image processing... then trail and error is your friend and the option to do differently is always precent and you always has an option to evaluate the different options against each other.

So to me it's a bit nuts to, for example, grade footage render and then key, or remove noise render and then grade, or opposite, grade and then remove noise.

My advice is to keep it all in one batch tree and move the nodes around / build different branches and try not to do any sub renders then you will be able to evaluate whats best for each shot / situation and also build up your skills as you see all the options pretty much instantly.

Where as jumping in between programs with subrenders in between will keep you wandering off in all the wrong directions.
 
What program are you using to keep everything in one node tree? Thanks!
 
Hi!

We do it on just the selected source footage after the editing is done (editing based on the R3D files).

This way we keep the material to run through Neat Video small, but have cleaner shots during the entire, often VFX heavy post.

If we need to match CGI elements with the shot, its essential to match the level of noise (or not noise) that the final product should have.

Removing the noise later, e.g. after the grading, would mean to me that we may alter the entire look once again - which I would strictly deny on.
The colorist wants to see the final product. Not less, not more.

Cheers,
Axel
 
pinned!
 
VFX elements have none or a different noise pattern (from GI for example). So it (noise reduction or noise matching) needs to be done before compositing. If only the CGi elements become a match grading things might be easier to decide.

Before or after grading, depends - but mostly - for me it is before, as an extreme noise can be very colorful if lifted in grading. To pull then the noise later might change the appearance too much.

If the noise/grain of all shots is similar and no CGi or other compositing effects are wanted, I think it is up to the "post" when to do. However, to say it in general might be difficult and not advised.

(If you have a "noise-print" from a gray-card [e.g., for neat video] - any alteration of the material would distroy or limit the usability of the "noise-print").

I think Pro Colorist might answer this - how they love to pull keys/qualifiers - with noise or without. Perhaps that would answer it much better.

P.S.: I try for each production a different way, as I teach people, to gain more experience. In the last production (LAndsCApes) I worked from 5K, primary correction, then noise reduction and sharpening. Applied the look I was after. From there: export to ProRes444 in 4K and ProResHQ HD. Edit in HD (for BluRay and as Proxy for the 4K.) Not my preferred way so way (as it locks too much), but so far the fastest after editing for me. Anyway, noise pulled in 5K is certainly my favorite way so far.
 
Last edited:
it's totally different to be working all alone on your own stuff and having to deal with multiple artists and moving files around.

if i have an fx plate you better believe im gonna do nr before i apply affects, at least for viewing purposes. as for green/bluescreen, you get cleaner keys out of cleaner images - simple stuff - i use neat video. if i'm compositing cg elements i'm dealing with nodes in nuke, so i can toggle nr while i'm working, but then i like to match the noise pattern on my fx elements, then reduce everything altogether as a part of coloring, although sometimes i will do cc/nr in nuke as a part of my compositing for certain situations.

if i have a lot of noise you better believe i'm at least previewing NR when i'm coloring as it hugely affects color perception. where i put it in the chain really depends on what i'm doing to the shot, but i don't do transcoding or anything like that and always work with original files all the way through if at all possible.
 
One big thing about NR is that it affects the downstream video encoding for distribution codecs.

Sometimes adding or leaving original noise may seem making color perception better, such as smoother gradients.
But sometimes the opposite happens after encoding to e.g. H264 or other codecs, as the overall noise significantly hits the data rate. Consequently the encoders may apply their own, quite unpredictable noise reduction with results looking even worse.

Making the perfect picture always requires one to look at the results in distribution formats.

Cheers,
Axel
 
You must log in or register to reply here.
Back
Top