Still have flicker after Denoising with AviSynth

[unclescoob]

V-dub's temporal smoother is usually pretty good. I find it superior to avisynth's temporal soften, which is supposed to be it's equivalent (?)

Try either one of these options:

1. Turn off NV's temporal smoother completely, and use Vdub's temporal smoother AFTER NV in the filter chain. Use a setting of 4-5 on the temporal smoother. Anything after 5 will definitely result in ghosting and/or blur. I find that the result is that NV takes care of all the spatial, chroma noise, and V-dub's temporal smoother takes care of the rest, when placed in that order in your filter chain.

2. Use V-dub's temporal smoother (set it at 5 frames instead of 3. this will go veeeery sloooww, but will take care of more noise and not kill your details). Then add V-dub's temporal smoother right after, but at a 1-2 setting. It will take care of the rest of the flickering that neat's temporal filter didn't catch.

Neat's temporal smoother doesn't appear as effective in and of itself, but that depends on the noise and your source. Keep in mind, the goal is to kill as much noise, but avoid the ghosting and blending. At least Neat's smoother doesn't cause any of those problems (if used correctly).

[sanlyn]

Isn't the Neat Video temporal filter a pretty dumb frame blend?

i.e. no motion compensation, no intelligence, and only basic conditionality? Basically equivalent to AVIsynth's temporalsoften()?

MVDegrainMulti for general noise - as used in the film restoration script - is usually far better.

btw, RemoveDirtMC (also as used in the film restoration script!) is near-magic for single frame junk.

Cheers,
David.

Explaining how NeatVideo works seems to be guessing at it. Its makers haven't revealed much about NV innards. It's apparently designed to work mostly on the noise sample the same way audio filters have to sample a bit of stable audio to determine what's noise and what isn't. As for motion compensation, I've posted several samples in the forum that tells me NV uses some form of motion compensation. I also tried the AviSYnth plugins you mention to target the flickering objects in the video discussed here; they didn't touch the target at all -- but, then, neither did NV or MCTemporalDenoise used alone (I don't think either filter is designed to work on those fluttering booboo's in the sample video).

A sample of some godawful/bad capture/damaged tape video fixed with a combo of Avisynth and NeatVideo was posted here: https://forum.videohelp.com/threads/322848-Scripts-and-filters-for-an-Opera-%28sample-v...=1#post2097410 along with a script that uses mvTools before feeding the results to NeatVideo. You can play with any filters you like on the original source MPG and you'll need some pretty stiff settings to see results. On the other hand, I don't know mvTools all that well: maybe MVT has a function that could address the specific problem of objects changing shape while a camera pans a scene, but no suggestions that I found here or in Doom9 came anywhere near fixing it. I'll be the first to admit that the sample shows strong filtering, but anyone who wants to tackle it is welcome to do so. The same thread has other posts showing that NV can clean many motion problems and compression artifacts that other plugins seemed to ignore.

So, I don't know why some users don't get results with NV (or succeed in destroying video with its defaults), but I've done a lot of fiddling with it and seem to be getting somewhere. This doesn't mean (obviously) that someone couldn't get better results another way. It doesn't mean NV is a cure-all either. For most video in more decent shape, I don't use it.

Anyway, the combo of DeFlick and NeatVideo worked in the wrestling sample. Nothing else I tried would fix the target noise. And I tried plenty, for 3 days.

[jagabo]

Neat Video's temporal filter is separate from its spacial filter. I believe the temporal filter isn't much different than VirtualDub's Temporal Smoother. I don't think is uses any inter frame motion compensation. The spacial filter probably uses fourier analysis to analyze intra frame noise (hence to need to train it with a noise sample) then uses what it learned from that sampling to remove noise in each frame.

If you use Neat Video's spacial filter after other filters you need to train it with a sample after those filters.

[unclescoob]

If you use Neat Video's spacial filter after other filters you need to train it with a sample after those filters.

Jagabo, thanks for confirming that. So my previously-posted method (as also mentioned by sanlyn), of applying temporal filtration in avisynth first, before sending to v-dub for Neat's spatial cleansing is a good idea then?

by the way, am I the only one who deems it a pain in the ass to work with NV from avisynth, and easier to do the two-step method?

[jagabo]

So my previously-posted method (as also mentioned by sanlyn), of applying temporal filtration in avisynth first, before sending to v-dub for Neat's spatial cleansing is a good idea then?

Whether you should use other filters before or after Neat Video will depend on the particular video and filter. But if you choose to use Neat Video after some other filters you must train Neat Video after applying those filters. Otherwise it will be removing the wrong noise. That's the whole point of training Neat Video.

by the way, am I the only one who deems it a pain in the ass to work with NV from avisynth, and easier to do the two-step method?

Using any VirtualDub filter from AviSynth is a kludge and a PITA. All filters that need two separate steps (ColourLike(), for example) are a bit of a pain in AviSynth. But they really should come out with a native AviSynth version of Neat Video.

[unclescoob]

What is a good program that I could use to mask certain parts of video? For instance, suppose I wish to apply stronger temporal filtration to the background of a scene only, and apply a different level of filtration to the moving object (to avoid ghosting).

[sanlyn]

Neat Video's temporal filter is separate from its spacial filter. I believe the temporal filter isn't much different than VirtualDub's Temporal Smoother.

Neat Video's temporal filter is much cleaner (neater??) and smoother with with fades in and out. VDub's temporal smoother can make them look grainy, clumpy and ugly.

[sanlyn]

What is a good program that I could use to mask certain parts of video? For instance, suppose I wish to apply stronger temporal filtration to the background of a scene only, and apply a different level of filtration to the moving object (to avoid ghosting).

After Effects.

[sanlyn]

[QUOTE=jagabo;2127546]

But they really should come out with a native AviSynth version of Neat Video.

Good question. Question #2: why does NV require RGB input but works internally in YCbCr? Actually, it works in more than one colorspace depending on input. Their manual sez they use something else if you specify you're working in grayscale. IMHO I find NV visibly more effective and faster working outside of Avisynth, but I haven't had that much time to compare the two methods.

I think unclescoob mentioned a "two-step" with NeatVideo. I do it in one step, after running a short test with the AviSYnth filters, testing an output patch for NeatVideo from the Avisynth results. Then I open VirtualDub, load my avs script, check that Avisynth is running as planned, then load my NeatVideo filter in VirtualDub; then I start the avs again with NeatVideo loaded, in one step. Configuring NV's settings in Avisynth is, I agree, a PITA. And readjusting it in script is even more painful.

[jagabo]

I think unclescoob mentioned a "two-step" with NeatVideo. I do it in one step

It's two steps in the sense you must first sample the noise in your source, then you filter it. You would have to do this even if it was an AviSynth filter.

[unclescoob]

I'm really trying to load an avi sample of what i'm talking about here, and I keep getting a message stating that the file is too large to process. But it's a mere 15 second huffyuv, so it's not that big. Why can't I load it?

[2Bdecided]

Whether you should use other filters before or after Neat Video will depend on the particular video and filter.

You should usually use NeatVideo before any external temporal noise reduction. That's how NeatVideo works internally - spatial then temporal (for the reasons you explained).

It can be a bit of a pain though. I like NeatVideo best for removing noise which has some kind of pattern to it, and mvdegrain for truly random noise. It's tricky to balance them and avoid ending up with plastic. NeatVideo easily generates plastic! I also hate the way mvdegrain leaves noise around moving areas/edges, while removing it perfectly on stationary areas. I know why it does that, and it doesn't look too bad when watching the video normally - but it looks horrible when you're comparing single frames to tweak settings. I like the way NeatVideo can leave in some high frequency noise (if you wish) - I often do that. It must be possible in AVIsynth by saving the noise and re-adding some of it (filtered) but I've never done it. I just cheat and add new grain if needed!

IME, mvdegrain doesn't remove real stationary details. NeatVideo will if you're not careful.

Cheers,
David.

[sanlyn]

I'm really trying to load an avi sample of what i'm talking about here, and I keep getting a message stating that the file is too large to process. But it's a mere 15 second huffyuv, so it's not that big. Why can't I load it?

The online load limit refers to file size in MB, not playing time. The forum limit is something like 9 or 11 MB (correct me on this, folks). 15 sec of AVI is likely around 125 to 150 MB.

[unclescoob]

So how can I load a sample clip? I've never done that here before.

[unclescoob]

Come on, After Effects costs an arm and a leg and has a ton of features that I will never EVER need! All I want to do is be able to mask different parts of a frame for proper filtering. Can't Sony do this? Or is there anything that I can acquire that strictly dedicates itself to masking and maybe a feature or two, but nothing else? I'd hate to spend 300 and somewhat dollars on a program that allows me to add fireballs falling from the sky....when all I need is to mask.

[unclescoob]

[QUOTE=2Bdecided;2127725]

You should usually use NeatVideo before any external temporal noise reduction. That's how NeatVideo works internally - spatial then temporal (for the reasons you explained).

2b, you just gave me an excellent idea. This particular episode (it's a cartoon). I'm going to filter the raw clip with Neat (with low temporal filtration settings), and then send it to Avisynth to add the Temporal Cleaner, which is a very good filter. What do you think?

[sanlyn]

Better: use Avisynth YUV before Neat Video RGB. Avoid colorspace conversations.

[2Bdecided]

Better: use Avisynth YUV before Neat Video RGB. Avoid colorspace conversations.

But they're irrelevant in the context of putting processing in the wrong order. Better to process in the right order with conversions, than the wrong order without. That's assuming the right order is, in itself, significantly better than the wrong one! Who knows in this case.

FWIW, IME, colour space conversions are also nearly irrelevant when you're dealing with a tonne of high frequency noise (unless you really are removing it all) because the noise dithers the re-quantisation and so hides the errors/banding/etc. Just use matrix="PC.601" both times to prevent clipping when converting.

If you convert after a strong effective temporal smoother, then you might see issues, since the source may be "too" clean then.

Cheers,
David.

[jagabo]

What is a good program that I could use to mask certain parts of video? For instance, suppose I wish to apply stronger temporal filtration to the background of a scene only, and apply a different level of filtration to the moving object (to avoid ghosting).

For still images you can create an alpha mask with a paint program, then use that mask to overlay two different filtered versions of the image.

src=WhateverSource()
mask=ImageSource()
v1=HeavilyFiltered(src)
v2=LightlyFiltered(src)
Overlay(v2,v1,0,0,mask)

Or you can try using MaskTools to make a mask:

src=WhateverSource()
mask=mt_motion(src).Blur(1.0).Blur(1.0)
v1=LightlyFiltered(src)
v2=HeavliyFiltered(src)
mt_merge(v2,v1,mask)

The problem with this is that just the edges are masked. It would be better to use McTemporalDenoise() which is fully motion compensated.

[unclescoob]

Isn't McTemporalDenoise one of those filters that isn't really a plug-in, but a script which requires ninety thousand other dll's? Or am I confusing it with TemporalDegrain?

[jagabo]

Isn't McTemporalDenoise one of those filters that isn't really a plug-in, but a script which requires ninety thousand other dll's? Or am I confusing it with TemporalDegrain?

Yes: http://avisynth.org/mediawiki/MCTemporalDenoise

[unclescoob]

Ok...I have no idea what to do with this. Can you just give me a quick and dirty method to load this? I will read the document, don't get me wrong. But I tried this filter before and loaded all the other filters. Can't get it to load in avspmod. Can you help me out with a simple script, and I'll tweak it from there? Jagabo, is this filter worth the trouble with heavy noise/flickering?

[jagabo]

Simple example:

AviSource()
ConvertToYV12() # if necessary
McTemporalDenoise(settings="high")

[unclescoob]

Aaaah, got it. Thanks. I guess if I use this, I wouldn't necessarily need to add NV to it, since this is a spatial-temporal cleaner.

[sanlyn]

One way to find out: try it both ways, then both together. If it's fairly clean to begin with, either should do the trick. MCTemporalDenoise on 'high' will be pretty slow. For a halfway decent video, medium might do it. You won't know until you try. Test a few frames first, see what you get. That's the only way to really get a handle on what these filters can or can't do.

[unclescoob]

You know, I gotta say: This is actually the first time I have an ongoing pleasant, helpful, "feel-like-I'm-getting-somewhere" experience at videohelp.com! I really like this thread! Thanks guys, I mean it. I feel like we're all collaborating here.

Ok, enough Danny Tannerisms. I'll give this a shot either tonight or tomorrow evening.

[sanlyn]

Some other samples of the difference between applying certain plugins in Avisynth and hitting its output in VirtualDub, in one step. I asked this question earlier in another thread, no answer followed.

These links are direct-play to a horribly damaged tape and equally horrible capture directly to DVD. The original VHS is lost, and wasn't mine. So the seriously flawed DVD capture is all there was to work with. I've learned a few things since the first go-round with this video and ultimately used different filters to fix it (sort of). Clicking the links will play the video in 4share's online player. No Audio.

A01_Wk50_TestA.mpg is a piece of the original DVD. It's a fade-out from one scene and a fade-in to the next scene. It's the result of applying YUV color correction in Avisynth, with no added filters. The frame size is the original 352x480 displayed at 4:3.

A01_Wk50_TestA.mpg
http://www.4shared.com/video/UICowpoW/A01_Wk50_Test

The same Avisynth color correction, this time with FFTD3 in temporal mode, strong settings:
http://www.4shared.com/video/WDWXueN_/A01_Wk50_TestA_FFT3D.html

The same Avisynth color correction, this time with FFTD3 as above, and Neat Video applied on the output in VirtualDub, in a single step:
http://www.4shared.com/video/qWXNLK0d/A01_Wk50_TestA_NeatVideo.html

The same Avisynth color correction, this time with nothing more than VDub's temporal smoother on output, set to 5:
http://www.4shared.com/video/nrRidZCR/A01_Wk50_TestA_Temp5xs.html

The temporal smoother version is the cleanest, even if admittedly it still looks like crap. But why the difference between applying this simple filter on Avisynth's output in one step, as opposed to saving the clip to RGB in one step and then running a second step on the saved clip with a simple temporal smoother?

[unclescoob]

Can I cut my mpeg in segments using dgindex? I would still like to post my sample clip here, and the avi is obviously too large, even in seconds-long clips.

[sanlyn]

Can I cut my mpeg in segments using dgindex? I would still like to post my sample clip here, and the avi is obviously too large, even in seconds-long clips.

I never tried editing with DGIndex. Two popular MPEG editors (free):
Mpg2cut2
AviDemux

I use TMPGenc MPEG Editor. Not free. They have a trial version, though.

I don't know whether or not they all work in a 64-bit OS if you're using one.

[sanlyn]

Come on, After Effects costs an arm and a leg

True. I recently bought an earlier version (CS3) for about half price. It wasn't an arm and a leg, but it was at least an arm. That was 2 months ago, I'm still learning the interface(!). But from what I read and see it's the best around, short of lab equipment that costs even more.

All I want to do is be able to mask different parts of a frame for proper filtering. Can't Sony do this?

Possibly. I installed the v9 Vegas Pro trial a while back, but it was limited in certain respects for my purposes. The SONY site has some tutorials and the user guide, and I've seen some clever Vegas Pro masking tricks posted on UTube.

Or is there anything that I can acquire that strictly dedicates itself to masking and maybe a feature or two, but nothing else? I'd hate to spend 300 and somewhat dollars on a program that allows me to add fireballs falling from the sky....when all I need is to mask.

I don't know of anything that will just mask, period. Maybe someone here has insight into MaskTools. But I know it to be a hassle to learn, and I believe I've seen it posted that area masking with it is difficult if not impossible. Then there's free Blender, but it would take months to learn its programming language for special techniques - even if it's possible to do what you want.

I agree, this kind of high-end software is priced outta sight and had me out of the loop for years. I'm still using Photoshop 5.5 (that's from the previous century, if you haven't heard of it)! You can do a lot more with AE and Vegas, not just add fireballs or extra noses. If you know someone in college or something with a student i.d. you could get the student packages cheap (relatively). But the learning curve is in weeks, not hours.