View Full Version : DeInterlacing filters
Rodger
23rd March 2008, 22:13
Well when I learned scripting and I had my harddisk full of stuff I had not much time to look out for several filters to try out.
So I took KernelDeint(order=0)
No idea of good or bad...it just worked. :rolleyes:
SOO...This question may not be new, but I didnīt see a matching thread (for some time).
Please name and categorize deinterlace filters of your choice.
1. Fast
2. best of both worlds (Iīd choose this one)
3. VERY HIGH Quality (Best quality)
Since there are so many filters Iīm sure some of them are outdated and shouldnīt exist anymore.
microchip8
23rd March 2008, 22:16
I'm on Linux so...
1) MPlayer's Linear Blend and FIR Lowpass 5
2) Yadif
3) Yadif + mcdeint
Guest
23rd March 2008, 22:51
78 posts and you're just blowing off rule 12 without a care in the world?!
Rodger
23rd March 2008, 23:19
Iīm not asking for whatīs best!!!
Mainly I was asking for improvement.
Iīm asking for sugguestiones...if you take Rule 12 that way...Iīd have to test ALL deinterlace-filters myself.
What Rule12 can mean is not to be able to use the knowledge of others. You know that.
78 posts and Iīm still a newbie in aviscripting and especially in itīs variety of filters and plugins.
YOU are VERY experienced...is there a way to ask for your opionion in this case?
/EDIT: Found out that MeGui shows higher FPS for encoding but takes longer than KernelDeint....how comes?
I use it yadif(mode=1,order=1) WRONG?
Guest
23rd March 2008, 23:26
Please name...
3. Best quality
Iīm not asking for whatīs best!!! Uh huh. Sure. Please take further discussion of rules interpretation to PM. Thank you.
You can just ask what people prefer for quality, speed, and a good combination of the two.
You should be aware that different kinds of material and different goals for it demand different treatments, and there is no universal prescription, especially when you ask about "best quality".
I don't do "encodes", but only double rate deinterlacing of short sports clips which I then slow down to half speed with AssumeFPS(). For that, sbdeint() works just fine for my needs.
Rodger
23rd March 2008, 23:36
You should be aware that different kinds of material and different goals for it demand different treatments, and there is no universal prescription, especially when you ask about "best quality".
Oops...got me there! I didnīt know to actually make any difference if Iīm deinterlacing xvid or mpeg1, or mpeg2. If I did get you right this time.
There are two things Iīm trying to do...
deinterlace h264 material from DVB-S2
deinterlace mpeg material from DVB-S(1) mostly videos from mtv (544x576 pixels)
You actually would use two different deinterlace-filters?
Aaahh...I think you are totally wrong about me and what capabilities I have :confused:
I would appreciate the easy way...like froggy1 understood me directly...no discussion about who has more knowledge. (hat would be you :p)
Guest
23rd March 2008, 23:59
I meant that you might "deinterlace" the following kinds of things differently:
-Fast sports video
-talking head videos with associated static displays
-field-blended videos
-phase shifted PAL
-hard telecined film
-soft telecined film
-hybrid material
etc. etc.
You have to know what your source material is before deciding if you even need a deinterlacer per se, versus IVTC, field matching, unblending, etc.
If you do indeed need a deinterlacer, then are you trying to make single rate or double rate output? Are there lots of static areas that you want to preserve? Do you care if you get some minor artifacts from some approaches?
There's just no simple answer.
I suggest that you post a link to a single clip and ask for suggestions about it. You'll learn a lot. Then if the approaches you learned fail for another clip, then post that one. Soon you'll have a full toolbox of techniques and know when to use them.
Rodger
24th March 2008, 00:28
How do you see your "KernelDeint" in comparison to yadif, or itīs derivat sbdeint?
Quick tests on my own showed...sbdeint is useless as the MTV-Logo there up right in the picture is almost shaking.
sbdeint is really faster than yadif, but to me obviously massively at the cost of "picture-control".
What makes your KernelDeint so fast in comparison...while itīs output isīn that bad in comparison.
/EDIT: Did I mention that the final files of sbdeint and yadif were about 0,5MB bigger in size? (63,1 to 62,6)
Guest
24th March 2008, 00:35
How do you see your "KernelDeint" in comparison to yadif, or itīs derivat sbdeint? As I said, I don't do any deinterlacing other than what I mentioned, so you'll have to wait for scharfis_brain to come along. :)
Quick tests on my own showed...sbdeint is useless as the MTV-Logo there up right in the picture is almost shaking. That's why in my post above I talked about "associated static displays". For my sports clips there are none, so flicker is not an issue.
sbdeint is really faster than yadif, but to me obviously massively at the cost of "picture-control". If your material has flicker-prone features, then use a deinterlacer that is optimized for such things. Again, that's my whole point about tailoring things to your material.
What makes your KernelDeint so fast in comparison...while itīs output isīn that bad in comparison. Well, it's fast because Leak optimized it. :) I assume you're using the faster LeakKernelDeint() rather than plain vanilla KernelDeint().
But seriously, it is a simple spatiotemporal kernel while yadif has more CPU-intensive processing. LeakKernelDeint() is a decent choice for a lot of material, but setting a threshold to preserve static areas can cause nasty artifacts for some fast-moving scenes, and it makes more jaggies on diagonals. That was one of my other points: are you willing to accept some artifacts?
Rodger
24th March 2008, 00:46
are you willing to accept some artifacts?
When it comes to the "fast" music videos...YES! Definetly....there is so much quality lost when sended by MTV...there is no deinterlacer in the world to compensate this :p
But when it come to my movie material from PremiereHD which is 1920x1088i (each movie about 8Gb of video + Audio-streams)....Iīd like to use a more versatile one...the allrounder.
Here is a test-video: http://rapidshare.com/files/101847092/neuron.mpg.html
Guest
24th March 2008, 01:22
But when it come to my movie material from PremiereHD which is 1920x1088i (each movie about 8Gb of video + Audio-streams)....Iīd like to use a more versatile one...the allrounder. You didn't post a sample of that.
Here is a test-video: http://rapidshare.com/files/101847092/neuron.mpg.html What is that? Is that the source or your processed result? I see no combing, it looks entirely progressive. Why do you want to deinterlace it?
Just in case you don't know... Just because the MEG2 is encoded in interlaced mode, it doesn't mean the *content* is interlaced (nonprogressive). You have to look at it. If and only if you see combing, then you need to do something.
Rodger
24th March 2008, 01:32
Yeah...itīs pretty good picture for 544x576x25i
But I do have to deinterlace, because my Samsung BD-P1400 wonīt accept interlaced h264 material.
There is no way to say: "this video is progressive" without deinterlacing, is there? (please say no!:eek:)
The other sample (h264 HDmaterial) is on itīs way to rapidshare.
Guest
24th March 2008, 01:40
Yeah...itīs pretty good picture for 544x576x25i I just told you it's progressive.
But I do have to deinterlace, because my Samsung BD-P1400 wonīt accept interlaced h264 material. So you are re-encoding your MPEG2 source to H.264? You need to decode the MPEG2 first. You will then have decompressed progressive video. Then do your H.264 encode progressively. You do not have to deinterlace.
There is no way to say: "this video is progressive" without deinterlacing, is there? (please say no!:eek:) See above.
Rodger
24th March 2008, 01:52
For MTV-Videos
DGindex > d2v-files > avs-script > MeGui (raw x264 output)
As much as I know...all TV-stations send interlaced material...
so even If I didnīt see the need I thought I had to deinterlace.
oh my... :(
Here is the 1088i h264 material: http://rapidshare.com/files/101857703/Scary_Movie_4_cut.ts.html
To say it clear...the h264 material iīm uploading only needs decent deinterlacing, if you ask me. does the work-around work here to?
Guest
24th March 2008, 01:58
For MTV-Videos
DGindex > d2v-files > avs-script > MeGui (raw x264 output) OK, but did you encode progressively in MEGUI? If so, why would your player complain? You are still confused about the difference between interlaced *encoding* and interlaced/progressive *content*. You can encode progressive content in interlaced mode, but that doesn't make the content interlaced.
As much as I know...all TV-stations send interlaced material... But that's obviously not true, because you just posted progressive content. Hello???
To say it clear...the h264 material iīm uploading only needs decent deinterlacing, if you ask me. does the work-around work here to? It's not a workaround!
I need to see the content before I can tell you what it needs. But I can't do that because you've used the crappy host rapidshare, which gives me one download per day and then asks for money. Try megaupload.com.
Guest
24th March 2008, 14:37
It's a new day and I am able to download your HD stream.
It, too, is progressive (PAL film) and does NOT need to be deinterlaced.
TheRyuu
24th March 2008, 14:42
1.) yadif (or yadifmod with nnedi)
2.) yadif (or yadifmod with nnedi)
3.) rule 12
Rodger
24th March 2008, 16:13
:thanks: @ neuron
I learned a lot!!!
But some correction to the MTV videos...that sample I uploaded, the kylie minogue video, showed harldy or absolutely no need for deinterlacing...just two videos further there was "JACK JOHNSON - If I Had Eyes" and this one NEEDS deinterlacing I can tell, when I checked the result :(
==================================================
Now Iīm looking for resizing...but I think (and hope) this one I understood better right in the beginning.
1088"i" > 720p via lancosresize(1280,720)
544*576(i) MTV > 576p via lancos4resize(720,576)
Guest
24th March 2008, 16:26
Be careful resizing interlaced material that you want to deinterlace. You have to deinterlace it before resizing.
Zarxrax
24th March 2008, 17:48
I think the topic of this thread brings up an important point. Going a bit further than the original question though, I think maybe we should come up with a comprehensive list of deinterlacers, and list the pros and cons of each.
I've been following this stuff for years, and it still confuses me. There are tons of deinterlacers out there, some of which I have never even tried. Trying to find the right one for any particular job can be an exercise in futility, especially if you don't know what to look for. And many of the deinterlacing filters these days let you combine them with other filters, for instance 'clip2' or 'edeint' parameters in tdeint. Without a solid understanding of what each deinterlacer is actually doing, things like this can not be used effectively.
Something like this would probably make a good page on the Avisynth wiki. Is this a good idea?
Rodger
24th March 2008, 18:50
I think the topic of this thread brings up an important point. Going a bit further than the original question though, I think maybe we should come up with a comprehensive list of deinterlacers, and list the pros and cons of each.
I've been following this stuff for years, and it still confuses me. There are tons of deinterlacers out there, some of which I have never even tried. Trying to find the right one for any particular job can be an exercise in futility, especially if you don't know what to look for. And many of the deinterlacing filters these days let you combine them with other filters, for instance 'clip2' or 'edeint' parameters in tdeint. Without a solid understanding of what each deinterlacer is actually doing, things like this can not be used effectively.
Something like this would probably make a good page on the Avisynth wiki. Is this a good idea?
VERY APPRECIATED!
I didnīt know it would be THAT difficult or so extensive. Obviously this theme goes VERY DEEP and takes a lot of time to get experienced. So why not? Put all knowledge about DeInterlacers together into one comprehensive thread. Itīll make this thread a pool of know-how.
Rodger
24th March 2008, 18:53
Be careful resizing interlaced material that you want to deinterlace. You have to deinterlace it before resizing.
Sure thing...My script:
#LoadCplugin("I:\Video\Avisynth\yadif.dll")
AVCSource("Inside Man.dga")
#yadif(mode=1,order=1) #good deinterlacing
#LeakKernelDeint(order=0) #fast deinterlacing
#Crop(0,4,-0,-4) #manual cropping
#AddBorders(0, 16, 0, 16) #Aspect Ratio correction
#Autocrop(mode=0,wMultOf=16, hMultOf=16)
LanczosResize(1280,720)
Didée
24th March 2008, 20:17
Since you're capturing from a PAL music channel, there's another non-trivial problem ...
If you're working with a music clip that comes from a PAL region, then most times it is progressive, everything is good. Rarely they will be interlaced, and can be deinterlaced without much problems.
BUT, if you're dealing with a music clip that comes from a NTSC country (e.g. everything from the USA), then there will be a big problem. Those clips almost always will have been converted from NTSC to PAL by a converter box, and those usually use fieldblending.
These clips can not simply be deinterlaced, because of the blended fields. (Well, you can deinterlace them, but the results will be poor.)
Search a bit on "blended fields", "Restore24" and "MRestore", to get into the matter.
audyovydeo
26th March 2008, 18:13
I think the topic of this thread brings up an important point. Going a bit further than the original question though, I think maybe we should come up with a comprehensive list of deinterlacers, and list the pros and cons of each.
...
Something like this would probably make a good page on the Avisynth wiki. Is this a good idea?
I'll second that - it's a common problem.
Personally, having invested a considerable amount of time in TDeint , I'm not about to switch to another (OK, it's also the most complete deinterlacer out there).
A good comparison, anf fairly recent (think it's early 2007) is YUVsoft's :
http://www.yuvsoft.com/pdf/Deinterlace.pdf
I understand YUVsoft to be an offshoot of MSU.
EDIT : this is NOT actually the PDF I was looking for : last year they had a good comparison of several deinterlacers, TDeint, yadif and others included. Can't find it right now but havent' got much time to look.
cheers
audyovydeo
scharfis_brain
26th March 2008, 18:39
From what I've seen so far, YUVSoft usually does unfair comparisons to enhance their own products.
A very unfair comparison was YUVSoft's stabilizer against GUnnar Thalin's Deshaker.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.