nnedi2/nnedi3/eedi3

[Dogway]

I really like how nnedi3 evolved. I compared it to eedi3 which I once had it as the ultimate solution for strong AA and surprised myself on how nnedi3 outperformed it, which says a lot. Anyhow I observed that under occasional circumstances eedi3 was doing better. Maybe it is just due to the nature of the filter itself, but wanted to take it noted.

source(x2) eedi3 nnedi3

[hamsterstyle]

What advantage is there running NNEDI3 as a de-interlacer rather than tdeint()? What kind of math does it do differently? Does it still make decisions about where the combed areas are by using a motion mask? Or is it purely spatial?

Also,

Is there an advantage to running say:
Yadifmod(Edeint=nnedi3()) or Tdeint(edeint=nnedi3()) as opposed to just running nnedi3 by itself?

[SSH4]

Dogway i played with nnedi/eedi too. replace nnedi to eedi on daa(). and result made me happy too on some anime source. nnedi3(pscrn=0) made shrper result (on daa()) but eedi on it made more correct result.
so both plugins are really usefull

[Chainmax]

So, as of v0.9.4 the default etype is better for image enlargements and for deinterlacing we should switch to etype=1, right?

[huhuyaya]

if can be used to downsize~
just like:LoadPlugin("nnedi3.dll")
nnedi3_rpow2(rfactor=-2,cshift="spline36resize")
that will be great

[Bloax]

There's no reason to use something as hefty as an (artificial) neural network ("NN"EDI) interpolator for something that's easily handled by something like Bilinear and Lanczos.

And thing is, NNEDI, if memory serves, is indeed "Neural-Network Edge-Directed Interpolator"
With interpolation needed for upscaling, since we need new data to fill the gaps.

With downscaling, we already have all the data we need.
So yeah, use Lanczos/Spline, or hell - BilinearResize(x,y) for that.

[Portioli]

hello

i would like to do some kind of supersampling (1080p --> 4K) using needi3,
then Sharpen the image using LimitedSharpenFaster
and then downsampling back to 1080p using lanczos or spline64resize.

could you please help me write the avisynth script?

[PhrostByte]

Originally Posted by Portioli

hello

i would like to do some kind of supersampling (1080p --> 4K) using needi3,
then Sharpen the image using LimitedSharpenFaster
and then downsampling back to 1080p using lanczos or spline64resize.

could you please help me write the avisynth script?

LSF already super-samples the clip internally, so this won't be as brilliant as you hoped.

[SubPixie]

I'd be curious to see if it looks good anyway, don't hesitate to post your findings Portioli

[Pulp Catalyst]

just a quick confirmation,

Startrek DS9 (really weird deintelacing, many here probably already know that a lot of 90's scifi programs are difficult to handle)

anyway, for this particular source is there any better combination than

"Yadifmod+NNEDI3"

for the record "QTGMC always crashes Xvid and x264 when using with Staxrip and MeGUI"

i'm just after confirmation if yadifmod+nnedi3 is as good as it get or near as, any feed back i would really appreciate it.

EDIT

DVD to Xvid/x264 (no resizeing, only cropping....MAX quality takes priority, but there is still an upper limit on file size i won't go over)

[SubJunk]

Have you seen the MEECH DS9 releases? He encoded all seasons at max quality.

If you still want to do it yourself, I recommend QTGMC. I know you said it crashes but I use it with MeGUI all the time with no crashes.

[Pulp Catalyst]

when you look at each episode, the part where the defiant is just about to vanish in the background (intro credits), you will see some savere interlace problems (most deinterlacers have trouble with this section because it's so small), i have tried many..... sources shall we say... and all suffer the same issue.

I have just installed Xvid4PSP, has QTGMC built in, seems to work flawlessly with this program..... go figure.

seeing what it can do now, just running quick test and will compare with yadifmod+nnedi3

downside to Xvid4PSP though is having to put in

SetMemoryMax(512)
SetMTMode(5,3)
......
SetMTMode(2)
......
......


on every episode, i have asked in forum, there is no way around it...... the template system is very good and mature in development in Xvid4PSP, however not so good when wanting to insert certain lines of code permanently on all future projects (like doing DVD episodes)

if i recall, i also believe that i found it very hard to get 100% cpu usage when doing Xvid encoding to which was annoying, a symptom that Staxrip and MeGUI did not have with the above additions to the avisynth script.

i did try for sometime to findout why QTGMC crashed in MeGUI and Staxrip (although it was the encoder that crashed in MeGUI, however in Staxrip, it was the program itself that stopped responding), tricky to work out, encoder crashed in one, but application crashed in the other????

gave up in the end however....... i put the problem on the backburner (meaning will deal with the issue once and for all when i do my next round of PC tower upgrade cycle), what tests i did do however i realized quite quickly that my tripple core could not be used on large scale jobs with QTGMC.... just not realistic when talking about 24 x 7 episodes LOL

[Chainmax]

I have some homemade footage that shows artifacting on the right side of the frame, which I am trying to antialias away. The filterchain is as follows:

SetMemoryMax(900)

SetMTMode(5,4)

MPEG2Source("X:\wherever")

SetMTMode(2)

a=last

b=a.Crop(704,0,16,576,align=true).nnedi3_rpow2(rfactor=2,nsize=3,nns=4,qual=2,pscrn=4,cshift="spline36resize").nnedi3_rpow2

(rfactor=2,nsize=3,nns=4,qual=2,pscrn=4,cshift="spline36resize").Spline36Resize(16,576)

Overlay(a,b,704)

rest of the filtering

Distributor()

I am getting a crash that VDubMod attributes to an out-of-bounds error that might be originated in nnedi3. How can I modify this to be able to encode?


[edit] I just noticed that I was doing this previous to bobbing. However, that was not enough: in addition to placing it after bobbing, I had to sandwich the nnedi3 line(s) between turnright() and turnleft().

[plonk420]

is there a way i can NNEDI3 + EEDI2/3 while keeping 60fps? (e.g. nnedi3's field = -2)

i tried EEDI3(field = -2, sclip=nnedi3(field = -2)), (as well as 2) but it has the "poorly deinterlaced video bounce"

edit: and can this be done with just built in filters?

[NicolasRobidoux]

http://www.imagemagick.org/discourse...tart=15#p88094 (NNEDI3) VS http://www.imagemagick.org/discourse...tart=15#p88102 (Elliptical Weighted Averaging with slightly sharpened Jinc-windowed Jinc 3 with sigmoidal halo minimization).
P.S. I should have written "experimental" instead of "state of the art": "sigmoidization" as a halo minimization method with negative lobe filters is just over 2 weeks old.
P.S.2 Better sigmoidal result: http://www.imagemagick.org/discourse...p=88156#p88155

[wOxxOm]

second picture has more aliasing and blur with just a few better scaled small spots.

[pbristow]

Hey tritical, you seem to be the go-to guy for EDI solutions in AviSYnth... Have you ever considered doing an implementation of "Fast EDI-1.5"?
http://www.eie.polyu.edu.hk/~wcsiu/p...1569292991.pdf

The paper talks about it being both faster and more accurate for uscaling images by a factor of 1.5x (e.g. for going from 480p to 720p) than the traditional approach of using NEDI to go up to 2x, and then downsizing by 3/4. Could be a useful addition to the AviSynth upscaling armoury...?

[active1]

is the nnedi3 resizer better than ResampleHQ?

[pbristow]

Originally Posted by active1

is the nnedi3 resizer better than ResampleHQ?

In some ways yes, in some ways no... They're each tackling completely different problems with the resizing process: NNEDI fills in missing *detail* (sort of) to prevent staircase artefacts and broken lines, while Resample HQ works around the side-effects of non-linear sampling of the colour values of pixels. Which one you would chose to use depends on which is the bigger problem in the case you're dealing with.

Ideally we would have NNEDI-ised version of Resample HQ, (or to put it another way, a gamma-aware version of NNEDI3) to combine the benefits of both...

[active1]

Originally Posted by pbristow

In some ways yes, in some ways no... They're each tackling completely different problems with the resizing process: NNEDI fills in missing *detail* (sort of) to prevent staircase artefacts and broken lines, while Resample HQ works around the side-effects of non-linear sampling of the colour values of pixels. Which one you would chose to use depends on which is the bigger problem in the case you're dealing with.

thanx, but can you please explain what kinds of images that i should use NNEDI3 or ResampleHQ with?
for example, if i had a linear image and i want to resize it with gamma-correction then i should choose ResampleHQ, right? if it is, what about NNEDI3?

Originally Posted by pbristow

Ideally we would have NNEDI-ised version of Resample HQ, (or to put it another way, a gamma-aware version of NNEDI3) to combine the benefits of both...

that would be great