Denoise Filter settings and a few minor bugs?

Started by xtro, September 10, 2017, 11:10:12 AM

Previous topic - Next topic

xtro

Hello
I've recently returned to using Avidemux after many years absence and it is pleasing to see a stable 2.7.x branch. My initial observations have been positive. I would like a bit of advice from more experienced users please regarding my observations using Avidemux in WinXP and Win7 where I tested 2.7.0 version.  I'm reasonably well versed with Avisynth scripts, their usage, chain format and options. Here we go!

Fluxsmooth at default 7/7  is I assume the settings for FluxsmoothST which only works on progressive sources.


Quote from its readme 'Interlacing: Progressive or interlaced (FluxSmoothT), or progressive only (FluxSmoothST)'

The other option for the temporal only FluxsmoothT is the one which does both interlaced and progressive, so if I decrease the spatial settings  to O or can it go to minus -1?(if so one should be using the other version anyway!) does the script then switch to FluxsmoothT, therefore the script which Avidemux uses a port of can be used on both an interlaced or progressive source?

Similarly the other two ports of HQDN3D especially the HQ denoise settings I see are very high indeed( if I'm reading correctly?), if they are set at default. A more conservative 2:1:3:3 would be far more less to destroy a video or cause ghosting then the present spatial setting etc which to me appear quite high for moderate denoising. Also according to Hqdn3d it too is only used on progressive sources, but is the avidemux port version able to be used on interlaced too?

Also I noticed a few bugs in 2.7.0 build. Before any encode the systems XP/Win 7 appear to 'hang/freeze' for over a minute or so before the encoding starts be it a 1 pass or 2 pass. Checking using Control Alt Delete shows the warning 'this program is not responding!' message. Wait awhile and it starts encoding. This freezing is in any video or audio operation.

Another bug? I found was when you just output the audio say to mp3 the resultant file has no file extension and is thus unreadable. This necessitates you to enter the films system in Windows and adding the .mp3 so the file can be read. Is this a bug in AviDemux or a permission issue in Windows?

For future feature requests as deinterlacing is a minefield for most users would an auto interlacing pattern detection feature for a DVD source(the one in AutoMKV is the best I've seen, alongside leaving telecine and decomb options in Handbrake set to auto which effectively manages combing effects or risks of duplicate frames) which is hybrid or pure interlaced be implemented rather than leaving it to the user to choose the appropriate script and make a mess of it to ivtc it correctly?

Thanks for your thoughts here.

xtro

#1
Also experienced bugs with the auto cropping plugin in 2.70 build.  Autocrop at the start of the video you get for example two figures top and bottom, move your video along to a different part and the figures for top and bottom change. Any position in a video does this.

Also another bug using editing a segment A and B of video throwing out warning of  'wrong length' or 'output is unrecognised.' I saw this with mkv as an input using clips of around a minute in length and using 2 pass h264.

Jan Gruuthuse

Quote from: xtro on October 01, 2017, 06:31:28 PM>8 >8
Also another bug using editing a segment A and B of video throwing out warning of  'wrong length' or 'output is unrecognised.' I saw this with mkv as an input using clips of around a minute in length and using 2 pass h264.
Probably this has something to do with one of the streams inside the .mkv container. Would be helpful if you uploaded these 1 minute mkv each causing the issues:  'wrong length' or  'output is unrecognised'.

eumagga0x2a

Quote from: xtro on October 01, 2017, 06:31:28 PM
Autocrop at the start of the video you get for example two figures top and bottom, move your video along to a different part and the figures for top and bottom change. Any position in a video does this.

Works as intended / as coded. Autocrop doesn't scan the video. You give it a single image from the video, it calculates crop values based on this image only.

xtro

#4
I'm not disputing how it is meant to operate!

What I am saying is it does this: when you move the slider to a different video position(for a recheck) new crop figures are collected which are different numbers from the previous estimate. The crop figures should remain the same for the whole video length/resolution shouldn't they after initial cropping? The calculus or mode of seems to be inaccurate.

Re:Mkv contents/streams
I think it is only a glitch if the mkv contains a srt file. This does not occur on edited segments 1 minute or above without the srt included as I tested before to confirm.

eumagga0x2a

Quote from: xtro on October 02, 2017, 07:32:29 PM
What I am saying is it does this: when you move the slider to a different video position(for a recheck) new crop figures are collected which are different numbers from the previous estimate.

No, this is not the case. At least I can't reproduce such behaviour. The values are recalculated only when I press the "Auto Crop" button.

QuoteThe calculus or mode of seems to be inaccurate.

This is another topic, but the calculation really seems to be incorrect. At least flyCrop::autoRunV seems not to work right (black bars on the left and on the right added by the "blacken borders" plugin are not detected by this helper function).

xtro

#6
When I wrote 'Recheck' I'd intended to convey a new pressing of autocrop button along with moving the slider to a new position. I was not very clear there, but we are both on the same page now, as you concede that new figures are generated after the pressing of it(autocrop) each time on the same source.

Nemesis

Hello. Nice to be here. Keep up the good work.

Like xtro, I was not very satisfied either by the default settings for HQDN3D.
I used the 2:1:3:3 settings suggested by xtro and I got better results.

May I kindly ask xtro what settings he/she would use if he opted for a less conservative approach, i.e. a little more than moderate denoising?
I would like to ask other forum members too about their favorite HQDN3D settings.
Thank you in advance.

xtro

#8
Hello Nemesis thankyou for your interest regarding denoise.  The first two are spatial and the last two temporal with the filter you refer to. With the other version mplayer denoise 3D(much better denoiser!) there are 3 settings only for some strange reason, again set too high and at these settings will destroy fine detail of any source. Set these to: 1:2:3 for a very light denoise or for your slightly noisy DVD 2:1:3 to clean it up. I personally prefer vague denoiser based on wavelet action to clean up things or remove grain(mode 2) for black specks on film sources like you can use in Staxrip or AutoGK.

Now regarding M player denoise 3D HQ follow suit 0:0:3:3 for any clean dvd action based source or any video input which might have slight motion artifacting or 1:1:2:2 for a super light allround denoise on a very clean source. H264 acts as a denoiser itself with its action and tends to smooth these potential problems out, so I would never use this denoiser on any already clean source. I might use it on a slightly grainy source 2:1:3:3 upping the spatial as I go, if the effect is not enough, so 2:1 will become 2:2:3:3 or 3:2:3:3 etc.

For floating specks you can increase the temporal- the last 2 numbers to say 5:5, but this denoiser I find has very little effect on that kind of noise and you might introduce color banding or blacks dots cropping up elsewhere or ghosting. On a video input say where you want to further compress(resize) and avoid mosquito noise just use 0:0:2:2. It will ever so slightly soften the picture, but you won't notice it. Fluxsmooth set at 2:2 or 3:3, is good for this clearing up of artifacts too. You'd be better off with FluxSmooth which is set at its default 7/7, increasing to 10/10 for heavier grained sources. If you can help it avoid using a denoiser full stop.

All you need to learn about this denoiser HQDN.. and its settings are in the link below. Hope it helps!

https://mattgadient.com/2013/06/29/in-depth-look-at-de-noising-in-handbrake-with-imagevideo-examples/

Nemesis

Dear xtro, you are a treasure!
Most tutorials seem too vague about actual parameter values.
I like that you make different suggestions depending on the situation.
Your post is very informative and to the point.
Thank you very much!

xtro

#10
My pleasure. After many years of using avisynth and failing miserably at first, but later after having some limited success, it makes sense to help others as using these scripts and plugin filters are not an easy task to get used to for the beginner.

My preferred script(other useful software I'd used!) for a very light denoise with light sharpening is:

FFT3DFilter(sigma=0.5, bt=5, bw=32, bh=32, ow=16, oh=16, sharpen=0.4)


For those tricky vhs like sources/recordings you have, increase the sigma to 2.5 or 3.0 with sharpening set at 0.5, or use Fluxsmooth filter alone set at 15 spatial, temporal 15.  Fluxsmooth is good enough to be just used alone and will work well in most situations. However all of this depends on whether the source is progressive or telecined and you might have to decomb telecide first then decomb decimate the source in your filter plugin order for the denoise plugins of avidemux to work as intended as they are deemed to work on progressive only.

http://www.avidemux.org/admWiki/doku.php?id=tutorial:deinterlacing_video