![]() |
MipSmooth 0.9 released
I've built a new filter to try out some new approaches to smoothing.
MipSmoother is an attempt at reinventing SmoothHiQ and Convolution3D. MipSmooth was made to enable smoothing of larger pixel areas than 3x3(x3), to remove blocks and smoothing out low-frequency noise. It is my hope that this filter will be better than C3D at smoothing out flat surfaces and gradients. This should be good for cartoons/anime, but it actually also helps film footage quite nicely. Get it at: http://cultact-server.novi.dk/kpo/av...MipSmooth.html Requirements: * YV12, RGB and YUY2 support. * CPU capable of running Integer SSE code. (P3, Athlon and newer). * AviSynth 2.5 required. * AviSynth 2.5.2 recommended. The new presets are fairly untested, and only based on a few sample clips, so I'd be happy to recieve feedback and propositions for changes. |
:)
another great filter sh0dan, great "reinvention" as you wrote. MipSmooth will be adaptable in the new script, seems great. i'm thankfull for you accept my invitation and post this news! :D |
sh0dan,
can you post a script with MovieHQ parameters? is simple like: MipSmooth(preset = "movieHQ") :?: i want use MipSmooth in my test. can' t wait to see the result. thanks in advance. :) |
Trying out your filter right now sh0dan :wink:
-kwag |
Hi there,
Does SpaceDust has a version for 2.52, I can only see the version for AVI Synth 2.0x By the way, Do we need space dust in 2.52? |
i did a little sample using the default parameter
MipSmooth() the result is good and clear...i'm doing more samples now. :) |
I just did a quick test, running ToK's sampler ( the second 10% sampler), and without the filter it took 42 seconds. Then I took away TemporalSoften(2,7,7,3,2) and added MipSmooth(preset = "movieHQ").
Now it took 1 minute to encode the same sample 8O Quality of both samples is almost identical, but encoding time is more that 25% :!: -kwag |
Maybe its worth trying another setting like
MipSmooth(preset = "movieLQ", scenechange = 8) |
Well Kwag you right again try different settings
and eventhough with the default parameters file size is smaller and a bit faster temporalSoften its still about 20 sec. faster, Quality its about the same i guess. |
@jorel: Yes - that is the way to invoke the presets - should work just fine.
@kwag: It's a bit more complex than temporalsoften, so it will take more time. It should be able to give considerably better image quality than temporalsoften - mainly because it does spatial work. It should be compared to c3d. I'm thinking about adding parameters to limit the number of mip (downsampled) frames that should be generated for each frame - also I'm looking in to possible cache techniques - so speedwise we are not out of options yet :) @syk2c11: Use the LoadpluginEx dll (search here or at doom9 forums). @all: I'm very interested in suggestions, and comments on the presets (too strong, too soft, artifacts, etc) - this is why it is still not 1.0 - it still needs a bit of finetuning. |
Quote:
I'll play more with it :!: -kwag |
Thanks mate,works great! :D
|
Superb filter. May be a little slow, but that doesn't bother me. Definitely works a lot better than Convolution 3d! I just use the movieHQ or LQ setting and it comes out great. Seems to be a lot less noise without adding a lot of blurring. I was getting worried about not finding a filter that was close to Convolution since the develpment has been stopped, but now comes mipsmooth!!
|
true digitalize,
i did some tests and got good results with movieHQ. convolution is great and MipSmooth is better...little slow, maybe sh0dan will improve some more speed, will be cool. :wink: |
I just tested the filter yesterday, and something is going wrong I think... I tested it on a noisy full-screen 90 min. long Capture, and it took 15 hours to encode the movie at 480x576 resolution... :!:
I'm on Athlon XP 1800+... Is it really THAT slow or is something going wrong? |
I use the MovieHQ settings listed in the html file that comes with the filter, but I change the upsize and downsize both to bicubic or lanczos. Looks almost identical to the original. Amazing. I use CCE, and it does seem extremely slow compared to other filters. Since CCE is so fast I can live with it. The speed is comparable to FaeryDust or PixieDust.
|
I have tried this filter again, and the results are simply amazing 8O
The only problem is speed :!: Encoding speed is X5 slower (compared with the filters in the current script ), so right now, I would only use it on short films and on night runs :) I hope sh0dan can optimize the filter :) -kwag |
So what are your recommended settings kwag? where did you put it in the current script?`
I really don't care about time when it comes to quality, so please post your AVS script here... |
5 Times Slower 8O 8O 8O
Wouldn't that mean that a 5 Hour encode would be a 25 Hour encode??? |
Quote:
|
Quote:
I used: MipSmooth(preset="MovieHQ") -kwag |
Hey, hey Don't think that I'm bad at Mathematics :lol:
It's just such a big time Time difference, that I required some confirmation. Well, I think 5 Times slower is just too much. The current optimal Script is very developed, already. So for me It's not worth it to wait 25 Hour's for a result which would only slightly be better. And also the Tok Prediction for PAL is a reason. I don't want to wait 25 Hours for a result that is probably too big. |
Quote:
-kwag |
:!:
after a test with MipSmooth after STMedianFilter was needed 5 times more to encode and 15 seconds to encode each 27 frames, more or less 56 seconds to encode 90 frames... :!: |
I think it looks great being used alone. So I do not even use anything else like STmedianfilter. I just change the resizers in the filter to bicubic or lanczos depending on the source. I find Kwags Optimal Script is extremely slow on 720x480 DVDs.(I only put two movies on a DVD, so I never resize my video, trying to keep them at the highest res for my future HDTV.)Adding mipsmooth to the Optimal Script will just kill it. Besides a couple of months ago you guys were all diehard 2.08 "we will never leave it" , "space dust rules!!!"fans. :) Now you guys are into stmedian filter and 2.52, so this could just be another addition\change to KVCD filter group. :)
|
digitalize,
:) in my third post in the first page i did tests with MipSmooth and the results was great (like you posted) using: MipSmooth(preset = "movieHQ") and MipSmooth() but using in the Kwag script after STMedianFilter was too slow and the result was good too. resuming, the final quality is cool but take long time. do you have some idea to use MipSmooth in the script and turning it faster? ...will be welcome! :wink: |
You can take all the filters out (current script) and use MipSmooth alone. Now time your encode.
Now remove MipSmooth and add the current script again. Time your encode again. Result: With MipSmooth, it's still 4X the time to encode :wink: The filter chain that is in the current script is a better choice (for the time being), because we are able to "plug" filters for different conditions. Not to mention the "Adaptive" nature of the script, which takes advantage of the motion difference between frames. -kwag |
Site design, images and content © 2002-2026 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2026 Jelsoft Enterprises Ltd.