SmoothIQ too slow, alternative?

[angelyote]

Hi guys,

I'm encoding an anime and this one has a lot of 'edge noise'. I'm sure you know what I mean if you've done animes. Anyway, I tried using SmoothIQ and msharp and found the results to be really good. Unfortunately with SmoothIQ it would have taken around 4 days to encode my first pass in virtualdub using DivX5. /boggle

Is there something someone can recommend that does what SmoothIQ can do but faster? I was using Diam 11, Stength 200 with it.

Thanks,
Dave

[theReal]

As far as I know, SmoothIQ is a spatial smoother - so you could try avisynth's native Spatialsoften() - it's really good in the latest versions of avisynth (I'm using 2.06).

I'm using it in this combination for my Simpsons tv-captures:

temporalsmoother(4,1)
SpatialSoften(2,5,7)
temporalsmoother(4)

The result is really good, but it's also pretty slow (like all spatial smoothers...). Maybe it's a little faster than Smoothhiq, I don't know for sure.

[angelyote]

SmoothiIQ is different from SmoothHIQ

I'm not really great at what the difference between Chroma, Luma, RGB and YUV is, etc but this is directly from the SmoothIQ webpage.
http://www.fiction.org/www/smoothiq/

Smart Smoother IQ is a filter for VirtualDub similar to Donald Graft's Smart Smoother, but operates in IQ (chrominance/"colour") space instead of RGB. Luminance (Y) is left intact. It was designed to reduce the appearance of "rainbows" (cross-colour artifacts) seen in certain anime DVDs, such as AnimEigo's Bubblegum Crisis DVDs and Pioneer's Trigun DVDs.

The filters works really well but I found it to be so slow it's just not feasible for me to use.

Dave

[Acaila]

Well, a radius of 11 for any spatial smoother is quite insane. No wonder it's slow as hell. Try turning it down a few notches (3 or 5 work great normally) and you should see quite a substantial speed increase.

Ps. Can someone move this to the VDub (or maybe the Avisynth) forum, as this is not a general discussion question.

[Zarxrax]

The filter does not work nearly enough if you turn it down. I often set both settings to the max and would ofter wish to add a 2nd instance of the filter, however the time it would consume is just too much.

[angelyote]

Thanks for all your replies guys.

Lowering the diameter certainly makes it faster but also seems drastically reduce the beautiful results I was getting with it.

Dave

[theReal]

I didn't notice this filter was called smothIQ and not -HIQ...

...but...just a thought: avisynth's spatialsoften also does not work in RGB but in YUY2 colorspace. If want to leave the luma intact, just lower the luma value (to 1 or 0(?) if possible): Spatialsoften(2,1,8).
However, probably it's not faster with a very high diameter...

[ErMaC]

Tim's filter is the only thing you can use if you're working in Vdub only, but if you feed your file thru AVISynth, you can use any smoother by going:

MergeChroma(filtername(options))
This performs "filtername" on just the chroma channel.

Of course, you can also specify things like convolution3d to only work in chroma space:
Convolution3d(1,0,8,0,8,3,0)

[amni]

Since I work only with VDUB I cannot compare it with
all the noise reduction filters. Amnong those which
I tested this is the best filter to reduce the noise
of tiny chromatic distorsion and the the same time
preserving the important "shape details".

I think it preserves the "shape detais" by combining
two paradigms:

1
conforming to MAIN EDGES (edge detection is included),
2
preserving the luma (human perception build on luma
as main shape identifier, the chroma details are
enhancements for the perception).

I might add details later, but these details
are enough.

The only problem seems is that this project was not
finished (especially, maybe it could be
10 times faster).
May be because of low speed few peopleuse it.
The current 0.6 beta version of SMOOTHIQ is
slower then the WMV9 encoder (2 pass, complex).