MipSmooth v0.9 released

[sh0dan]

FYI .... there is something strange with very high spatial thresholds and low temporal threshold... currently investigating!

edit: It seems to be "intended operation" - Back to the drawing board - there is little use of it, if it actually increases noise in some cases.

[gizmotech]

Hmmm I've just tried to do some basic testing of mipsmooth, 0.9 in YUY2 mode, and unfortunately I can't seem to get it to work. When you said limited YUY2 support, how limited are we talking?
my avs:

LoadPlugin("C:\newfilters\mpeg2dec3.dll")
LoadPlugin("C:\newfitlers\mipsmooth.dll")
mpeg2source("bloodylongpath\episode5.d2v")
YV12toYUY2(interlaced=true)
MipSmooth(10,10,5,10,5,method="soft")

Now here's the thing, I can get ms to run fine in yv12 mode, however using the YV12toYUY2 command or the mpeg2dec.dll for avs 2.5 I cannot get it to start. It continually returns Unrecognized Exception! on the ms line.

I've updated every plugin I can think of(which is 2 in this case:(), avs is 2.5.2, and I'm running out of ideas. What YUY2 sources did you try it on? better yet, what are the base requirements for YUY2 functionality?

GizmoTech.

[Zarxrax]

Great work on the filter shodan! This looks like one of the more promising denoisers to come around in a while. However, from a quick test run that I did, it didn't hold up against deen + temporal cleaner.

Here was my results, all encodes done in xvid 100% quality.

Normal clip: 7.13mb
Looks a little noisy, nothing too terrible though.

MipSmooth with AnimeHQ preset: 6.88MB
Good compression gain, noise removal seems good, though the image is noticably less detailed.

MipSmooth with Anime LQ preset: 7.35MB
No compression gain here, in fact, just the opposite! Noise removal was far too much for this clip though, I'll have to try it out on a crappy source sometime.

Deen("a2d",5,5,10).TemporalCleaner(4,6) : 6.51MB
Largest compression gain. Also retained more detail than MipSmooth did. In the flat areas of the picture, it also appears that it *might* have smoothed them out better than MipSmooth did.


I'd also like to note, that on a scene that had a lot of detail and almost no flat areas, neither MipSmooth nor Deen caused any loss of detail that I could notice. It seems they only lost detail on scenes that had both flat areas and detailed areas.

[unimatrixzer0]

It seems mostly everyone has been testing this on dvd sources but has anyone tried it on noisy tv captures? It doesnt actually make my video look 'smooth', but leaves pixie dust/noise on my video instead. Same thing with c3d. Althought today I had a friend use c3d on a non noisy source and it turned out smooth so I'm assuming maybe mipsmooth (like c3d YV12 version) doesn't handle well with noise/grain? Tried all the presets and tweaking those with same results. Besides mipsmooth decomb is the only other filter I use so I'll assume its mipsmooth.

FYI .... there is something strange with very high spatial thresholds and low temporal threshold... currently investigating!

edit: It seems to be "intended operation" - Back to the drawing board - there is little use of it, if it actually increases noise in some cases.

I'll just wait and hope this is whats behind my problem till the next release.

[yaz]

Originally posted by unimatrixzer0
It seems mostly everyone has been testing this on dvd sources but has anyone tried it on noisy tv captures? It doesnt actually make my video look 'smooth', but leaves pixie dust/noise on my video instead. Same thing with c3d. Althought today I had a friend use c3d on a non noisy source and it turned out smooth so I'm assuming maybe mipsmooth (like c3d YV12 version) doesn't handle well with noise/grain? Tried all the presets and tweaking those with same results. Besides mipsmooth decomb is the only other filter I use so I'll assume its mipsmooth.
(almost) the same here. as my pc is sse incapable, i can't test it myself, but a friend o'mine gave it some chance (as he found c3d too 'harsh' for noisy tv capts) results are more or less the same as outlined above.
- in yuy2 there's no significant effect neither with some more 'agressive' settings. however, it does something, as at higher settings it started artifacting heavily.
- in yv12 it removes some noise, definitely, but the result is like described by unimatrixzer0 (what a name, i luv it :-) after some tweaking the noise dropped to an 'acceptable' level, but the effect is far beyond that of the fluxsmooth/tempsoft or dust/tempsoft combo.

anyway. when a filter gets suspicious for me i simply switch the thresholds to the upper limit, just to see whether it works or not (& this way i can see what kinda artifacts can be expected). but i can't do this with mip as my friend is not a 'lets tweak it for ever' kind :-)

that's all i've got so far. maybe help.

thx
y

ps offtopic : some 2/3d filters (say, fluxsmooth, deen,...) generates very strange color noise at higher thresholds. have u ever noticed?

[digitalman]

I got great results using the preset values for MovieLQ on a DVD movie. But I changed the upsizer and downsizer to bicubic. Did a lot better job than Convolution3d with the noise, and the picture was still sharp. My new choice of a denoiser for now.

[kilg0r3]

Sorry for this lazy question, but my exams ... and if that were not enough ...

Has anybody here done some testing on live action movies or is this an anime only filter?

[mf]

Originally posted by kilg0r3
Sorry for this lazy question, but my exams ... and if that were not enough ...

Has anybody here done some testing on live action movies or is this an anime only filter?
Look at the post above you :rolleyes:.

[drebel]

Great results with still scenes and...not so great with high motion scenes.Does the filter contain any motion detection algo ?The rest of the picture was perfect except of that high motion part which looked awfoul (blurry and with wrong(?)luma-chroma (something like a "negative effect")
Hope i'm wrong
regards,
george

[Defiler]

The "temporal" and "scenechange" parameters most definitely involve motion detection. Heh.
Sounds like you need to adjust the parameters. I'm not seeing anything particularly unpleasant as a result of using this filter.
What parameters are you using?

[MasterYoshidino]

this filter is one of the few spatial filters which is capable of removing vertical banding of colors. (thanks to the mipping subsampling). All other smoothers I tested fail to detect it as not detail.
Speed is something this filter lacks though lol.
Still need to look @ the frames with Photoshop to see what makes for better smoothing but preserve edges.

[Defiler]

Originally posted by MasterYoshidino
Speed is something this filter lacks though lol.I'm curious.. What MipSmooth settings are you using? With preset=AnimeHQ, I see barely any performance hit from enabling this filter.

[neily]

shOdan,

With v0.9, when the input is not YV12, an error message is thrown "Evaluate: Unrecognised exception!" whereas the previous versions provided the more helpful "Mipsmooth: Requires planar(YV12) input. It would be much appreciated to see the reappearance of the more informative error message.

TIA

[sh0dan]

It _should_ support all colorspaces. What are you feeding it (dimensions/colorspace/script)?

[neily]

shOdan,

No, really it doesn't seem to support YUY2. My original clip was a 352x288 picvideo MJPEG avi which when loaded by AVISource outputs in YUY2. Error with no conversion. No error with ConvertToYV12 preceding MipSmooth, and error again if preceded by ConvertToYV12 then ConvertToYUY2.

Also tried with an MPEG2 480 x 576 loaded by MPEG2Dec3. Outputs as YV12, and no errors with MipSmoth. Unfriendly "Evaluate: Unrecognised exception!" if MipSmooth preceded by ConvertToYUY2.

I am using AVISynth v2.52, and the latest version of MipSmooth. Loading via AVSGenie, VDub, and mplayer2 with similar results.

If you want more details let me know, or if it is working for you in YUY2 let me know and I'll do more fiddling.

Really, I don't mind if it doesn't work in YUY2, just a more friendly message would be helpful.

P.S. Seems to work fine in RGB24 and RGB32.

[sh0dan]

I cannot reproduce - could you post a script?

[neily]

shOdan,

Continuing to investigate, but have to be away from computer for a couple of days. Will get back to you. In the mean time, it seems that the errors occur in YUY2 if no parameters are used in MipSmooth. Can you reproduce this? If not I will do a more thorough investigation on my return.

[N_F]

Originally posted by Defiler
I'm curious.. What MipSmooth settings are you using? With preset=AnimeHQ, I see barely any performance hit from enabling this filter.
Mipsmooth easily doubles my processing time (not that I care much as I, as most, set quality over time)

[MasterYoshidino]

hmm theres good smoothing but there seems to be a temporal artifact or spatial artifact where backgrounds will end up flickerking.
http://membres.lycos.fr/etchianime/stuff2.htm
link to avi that shows the flicker artifact , look at the "lights" or the colorful disks as Sen/Chihiro walks up to the door. Maybe the more accurate way to describe this artifact is "going out of focus and then going back into focus", definately = bit rate that is not neccesary

[neily]

sh0dan,

I have had a bit more of a chance to investigate the conditions under which the "Evaluate: Unrecognised exception!" occurs (I assume now this is an AVISynth-generated error rather than a MipSmooth error).

It seems that it only occurs if the colour space is YUY2, the downsizer is set to reduce or bilinear, and the scale factor is set to less than 0.60. Thus it affects the default settings and the AnimeHQ and Anime LQ presets.

Is it just me, or is this a reproducible problem?

An example script is:

LoadPlugin("E:\Video\AVISynth YV12 Plugins\MipSmooth.dll")
AVISource("P:\PV Small 18.avi")
MipSmooth(downsizer="bilinear",scalefactor=0.50000)

[sh0dan]

Ah - could be a resizer limitation. It's probably the resizer crashing because resolution gets to low - I'll put in some min/max functions, that drops mips if they are too small.

[digitalman]

Is anyone else using this as their only filter? Is there gonna be a 1.0 released soon?

[sh0dan]

I'm working on some algorithm improvements, and some performance related options for 1.0 - but it's progressing (slowly nevertheless) :)

[digitalman]

Keep up the great work with MipSmooth!