AviSynth 2.5x and Convolution3D guide for AVI captures

[vhelp]

@vhelp

There is a VirtualDub filter called general convolution that works similar to how Convolution3D works but the VirtualDub filter comes with no documentation that I have been able to find so I never did ever figure out exactly how it works.

As for VirtualDub filters I prefer to work in the YUY2 colorspace since that is what most captures are to begin with. Thus my use of AviSynth, Convolution3D and Cinema Craft Encoder.

VirtualDub does an RGB conversion as does TMPGEnc

- John "FulciLives" Coleman

Thanks Fulci,

I do realize your point w/ respect to YUY2. The lesss "color space" conversion, the BETTER
And, I also believe that the final encode to MPEG (DVD for instance) would
come out better if there were no conversions to begin with, but sometimes
we (myself that is) use vdub for other reasons :P

Thanks again for your response back.
-vhelp 2550

[kenmo]

I followed your guide ...step and by step... it worked surprisingly easy. The output was very good....

Many thanks for taking the time and effort to put together this guide for us all here....

I would like to compare the output and time to a similar VirtualDub effort...

Anyways, my hat goes off to you....

Kenmo

[NamPla]

Thanks, FulciLives, for the terrific huide! I was desperate for something like this, despite what I read everywhere Avisynth ain't too easy to get started in, esp. when dealing with interlaced video.

Speaking of whjich, I have had a problem with 2 old VHS movies of mine. They are old b/w's, I think they were converted from NTSC to my native PAL. They are pretty bad quality, but clean up surprisingly great with Convolution 3D!!! The problem is, after burning to DVD and playing on my TV, I see horrible comb9ing effects when there's horizontal movement (not there before). It's not a field order problem (I checked).

So after experimenting, I simply inserted the Bob() command in my script HERE:

LoadPlugin("Convolution3d.dll")
avisource("C:\~\capture.avi", false)
Greyscale()
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Bob()
Weave()
LanczosResize(352,576)
ConvertToRGB(interlaced=true)

This seems to fix it. My question is, simply, is this the right place to put it? Should I be doing something else?

Cheers---

(PS Fulci's "House By The Cemetary" is one of my favourite movies! "Bob! Bob!"...)

[FulciLives]

@NamPla

So you are capturing PAL VHS into a PAL format AVI and then converting to PAL DVD format?

If so then you shouldn't have to deinterlace unless you are attempting something "funky" like PAL to NTSC etc.

- John "FulciLives" Coleman

[erratic]

So after experimenting, I simply inserted the Bob() command in my script HERE:

LoadPlugin("Convolution3d.dll")
avisource("C:\~\capture.avi", false)
Greyscale()
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Bob()
Weave()
LanczosResize(352,576)
ConvertToRGB(interlaced=true)

Avisynth assumes BFF by default. If your video is upper/top field first (TFF) you should use the AssumeTFF() command after AviSource. Try this script:

LoadPlugin("Convolution3d.dll")
avisource("C:\~\capture.avi", false)
AssumeTFF()
Greyscale()
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Weave()
LanczosResize(352,576)
ConvertToRGB(interlaced=true)

[erratic]

It looks like I'll have to correct myself. I ran a short test and in this case AssumeTFF() doesn't appear to change anything. Bad combing effects can occur if you vertically resize interlaced video, but I don't know if that's what's happening. What's your capture resolution, NamPla?

[NamPla]

So you are capturing PAL VHS into a PAL format AVI and then converting to PAL DVD format?

Yes, thats right.

AssumeTFF() doesn't appear to change anything. Bad combing effects can occur if you vertically resize interlaced video, but I don't know if that's what's happening. What's your capture resolution, NamPla?

I'm capturing at 720x 576.

The combeing looks like this on my TV set:-



Actually, its only this one old movie of mine that's doing it. All my other encodes have turned out GREAT using your guide, FulciLives! The improvement over the original VHS source is astounding!

[NamPla]

Hey I should mention that if I encode the raw avi WITHOUT any avisynth script, there is no combing problem! But the quality looks yuck like the original, ha.

It would seem the problem is somewhere in the separating of the fields, and the weaving them back together...??

Thanks.

[FulciLives]

Hey I should mention that if I encode the raw avi WITHOUT any avisynth script, there is no combing problem! But the quality looks yuck like the original, ha.

It would seem the problem is somewhere in the separating of the fields, and the weaving them back together...??

Thanks.

Sometimes old video tapes can cause problems with the field order. I'm pretty sure whatever is wrong here is a field order problem.

Try the AviSynth script like this ... just run a sample like maybe 1 minute and see how it turns out:

LoadPlugin("Convolution3d.dll") 
avisource("C:\~\capture.avi", false) 
Greyscale() 
SeparateFields() 
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0) 
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0) 
Interleave(evn,odd) 
Weave() 
DoubleWeave.SelectOdd()
LanczosResize(352,576) 
ConvertToRGB(interlaced=true)

I've added one line to this script. The line that reads DoubleWeave.SelectOdd() which is for bottom field video. The other way (leaving that line out) is for top field video but obviously that isn't working for you and since I know some "bad" tapes can actually switch the field order I would try it this way for bottom field.

So let me know what happens.

- John "FulciLives" Coleman

[NamPla]

Hi again, last night I captured a 1 minute sample from this movie & encoded with every possible combination of tweaks & settings. And guess what? It turned out fine first off ("TFF interlaced" in TMPGEnc, no "doubleweave" added in my script etc).

So...I've concluded that perhaps Erratic was right earlier when he said:

Bad combing effects can occur if you vertically resize interlaced video

Because, before I was using "Crop" in my script. With the samples I did last night, I didn't crop. Could this be the cause of it? Of course, when I cropped I was keeping it in even numbers of pizels (otherwise TMPGEnc wouldn't have opened it!). What do you think was problem here?

Thanks a lot.

[FulciLives]

Hi again, last night I captured a 1 minute sample from this movie & encoded with every possible combination of tweaks & settings. And guess what? It turned out fine first off ("TFF interlaced" in TMPGEnc, no "doubleweave" added in my script etc).

So...I've concluded that perhaps Erratic was right earlier when he said:

Bad combing effects can occur if you vertically resize interlaced video

Because, before I was using "Crop" in my script. With the samples I did last night, I didn't crop. Could this be the cause of it? Of course, when I cropped I was keeping it in even numbers of pizels (otherwise TMPGEnc wouldn't have opened it!). What do you think was problem here?

Thanks a lot.

Well the script that you posted ... which you said was giving you trouble with that particular source ... did not have any crop command in it. It did have a resize to 352x576 but as long as you captured at a height of 576 then the resize wasn't the problem.

Perhaps you cropped and resized instead of using the AddBorders command to fill in what you cropped?

Don't know since you didn't give us the crop values you used etc.

- John "FulciLives" Coleman

[NamPla]

Yes, sorry I didn't have my original Crop values anymore. But yes I think you nailed it there, it was my fault I wasn't adding the borders I think.

Basically I was trying to keep the aspect ratio when I cropped so it resized without warping the ratio.

I am thinking, instead of using "Crop" and "addBorders" , which I find fiddly, couldn't I just load the aviscript into TMPGEnc and use TMPGEnc's crop (which adds black borders automatically)?? This is much easier.

Thanks!

[FulciLives]

Yes, sorry I didn't have my original Crop values anymore. But yes I think you nailed it there, it was my fault I wasn't adding the borders I think.

Basically I was trying to keep the aspect ratio when I cropped so it resized without warping the ratio.

I am thinking, instead of using "Crop" and "addBorders" , which I find fiddly, couldn't I just load the aviscript into TMPGEnc and use TMPGEnc's crop (which adds black borders automatically)?? This is much easier.

Thanks!

Well I don't see why it is so hard to use the AddBorders command. For instance if you have to cut 12 from the bottom to remove noise from a VHS tape just put 6 black on the top and 6 black on the bottom with the AddBorders command.

It's very simple to do.

Resizing the height will cause problems unless you do it "just right" hence it is just easier to use the AddBorders command.

- John "FulciLives" Coleman

[NamPla]

Heh, you're probably right. I just get headcramps when I'm cropping 2 from top, 12 from bottom, 4 from left & 8 from right (for example), then having to add borders to get the picture centred again (without disrupting the aspect ratio)!?

Also I like to add an extra 8-16 pixels frame to compensate for overscan...

But hey, I'll go do more research. Thanks for all your help, and cheers again for the great guide!

[FulciLives]

Heh, you're probably right. I just get headcramps when I'm cropping 2 from top, 12 from bottom, 4 from left & 8 from right (for example), then having to add borders to get the picture centred again (without disrupting the aspect ratio)!?

Also I like to add an extra 8-16 pixels frame to compensate for overscan...

But hey, I'll go do more research. Thanks for all your help, and cheers again for the great guide!

Well if you learn to do it all in AviSynth then you can perhaps try another encoder otherthan TMPGEnc such as CCE BASIC which IMHO is a better faster MPEG encoder.

But you could do all the cropping etc. in TMPGEnc I suppose. However please note that I think (stress think) that if you use TMPGEnc and set the aspect ratio to KEEP ASPECT RATIO then TMPGENc assums the source width is 704 instead of 720 so that could cause a slight aspect ratio difference if you don't properly account for it.

So the "safe" way is to do it in AviSynth making sure your width is 720/352 and just select FULL SCREEN for the aspect ratio in TMPGEnc.

- John "FulciLives" Coleman

[NamPla]

Thanks again, FulciLives. Yes I will stop being lazy about it. CCE Basic looks good. But even with TMPGEnc, I am now encoding movies in about 4-5 hours...whereas before (frameserving with VirtualDub) it was taking 24+ hours! heh... So I'm happy at the moment.

(Also, I find the Convolution3D filter does everything a chain of VirtualDub filters does, and more!)

Thanks.

[FulciLives]

(Also, I find the Convolution3D filter does everything a chain of VirtualDub filters does, and more!)

Shhhhhhh !!!

Don't let SatStorm hear that

- John "FulciLives" Coleman

[SatStorm]

Actually I read it...

Convolution3D is not a simply filter. It does at once what 2 - 3 filters do seperate. And since it is an avisynth filter, it does it fast.

[FulciLives]

Actually I read it...

Convolution3D is not a simply filter. It does at once what 2 - 3 filters do seperate. And since it is an avisynth filter, it does it fast.

Come to the Dark Side ... it is your DESTINY ... you KNOW it to be TRUE :P

- John "FulciLives" Coleman

[NamPla]

Ha ha, you funny...

[NamPla]

Hi again, hope is OK to post another question here?

I been having total success now with this excellent VHS-to-DVD method!

But now I'm trying to convert an interlaced PAL video (some local band footage) to interlaced NTSC DVD-R to send to a friend in the U.S.

I'm wondering if it's possibly to combine these 2 scripts?---

LoadPlugin("Convolution3d.dll")
avisource("C:\Documents and Settings\Matt W\My Documents\My Video\capture.avi", false)
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Weave()
LanczosResize(352,576)
ConvertToRGB(interlaced=true)

LoadPlugin("SmoothDeinterlacer.dll")
AviSource("("C:\Documents and Settings\Matt W\My Documents\My Video\capture.avi", false")
SmoothDeinterlace(doublerate=true)
LanczosResize(352,480)
ChangeFPS(59.94)
SeparateFields()
SelectEvery(4,1,2)
Weave()
ConvertToRGB()

I want to do the PAL-to-NTSC conversion is this way because I don't really want to touch the audio.

Can I combine these scripts some way? I tried this:

LoadPlugin("Convolution3d.dll")
LoadPlugin("SmoothDeinterlacer.dll")
avisource("C:\Documents and Settings\Matt W\My Documents\My Video\capture.avi", false)
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Weave()
SmoothDeinterlace(doublerate=true)
LanczosResize(352,480)
ChangeFPS(59.94)
SeparateFields()
SelectEvery(4,1,2)
Weave()
ConvertToRGB()

(Don't laugh!). It didn't work, the result is jerky horizontal motion. I tried both TFF and BFF in TMPGEnc Plus. I also tried include the "DoubleWeave.SelectOdd()" command. Everything resulted in jerky playback on my DVD player. (My DVD player can play both PAL and NTSC, by the way). And I haven't been cropping at all.

Does anyone have an idea?

Thank you.

[erratic]

If your source is top field first you should insert the AssumeTFF() command. I've tried it and without that command the result of the PAL->NTSC conversion was jerky. You can insert the command after the AviSource command (certainly before the SmoothDeinterlace command).

Also after SmoothDeinterlace(doublerate=true) there's a double frame at the beginning of the file. You can test this by watching this script in VirtualDub:

AviSource("capture.avi")
SmoothDeinterlace(doublerate=true)

You will see that the first and second frame of the 50fps video are the same. This shouldn't happen, and can be fixed by adding Trim(1,0). So try this to see the difference:

AviSource("capture.avi")
SmoothDeinterlace(doublerate=true)
Trim(1,0)

No more double frame. But if your source is top field first both scripts will be jerky, so try this to fix that:

AviSource("capture.avi")
AssumeTFF()
SmoothDeinterlace(doublerate=true)
Trim(1,0)

The jerkiness is gone. Of course this only applies to top field first video. If your source is bottom field first, AssumeTFF should obviously not be used.

So try this script:

LoadPlugin("Convolution3d.dll")
LoadPlugin("SmoothDeinterlacer.dll")
avisource("C:\Documents and Settings\Matt W\My Documents\My Video\capture.avi", false)
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Weave()
AssumeTFF()
SmoothDeinterlace(doublerate=true)
Trim(1,0)
LanczosResize(352,480)
ChangeFPS(59.94)
SeparateFields()
SelectEvery(4,1,2)
Weave()
ConvertToRGB()

[erratic]

If your source is top field first you should insert the AssumeTFF() command. I've tried it and without that command the result of the PAL->NTSC conversion was jerky. You can insert the command after the AviSource command (certainly before the SmoothDeinterlace command).

Also after SmoothDeinterlace(doublerate=true) there's a double frame at the beginning of the file. You can test this by watching this script in VirtualDub:

AviSource("capture.avi")
SmoothDeinterlace(doublerate=true)

You will see that the first and second frame of the 50fps video are the same. This shouldn't happen, and can be fixed by adding Trim(1,0). So try this to see the difference:

AviSource("capture.avi")
SmoothDeinterlace(doublerate=true)
Trim(1,0)

No more double frame. But if your source is top field first both scripts will be jerky, so try this to fix that:

AviSource("capture.avi")
AssumeTFF()
SmoothDeinterlace(doublerate=true)
Trim(1,0)

The jerkiness is gone. Of course this only applies to top field first video. If your source is bottom field first, AssumeTFF should obviously not be used.

So try this script:

LoadPlugin("Convolution3d.dll")
LoadPlugin("SmoothDeinterlacer.dll")
avisource("C:\Documents and Settings\Matt W\My Documents\My Video\capture.avi", false)
SeparateFields()
odd=SelectOdd.Convolution3D (1, 32, 128, 16, 64, 10, 0)
evn=SelectEven.Convolution3D (1, 32, 128, 16, 64, 10, 0)
Interleave(evn,odd)
Weave()
AssumeTFF()
SmoothDeinterlace(doublerate=true)
Trim(1,0)
LanczosResize(352,480)
ChangeFPS(59.94)
SeparateFields()
SelectEvery(4,1,2)
Weave()
ConvertToRGB()

[FulciLives]

This is an example of an actual script I recently used to convert a 25fps PAL capture (which was from an interlaced PAL VHS) to NTSC DVD format.

LoadPlugin("Convolution3d.dll")
LoadPlugin("C:\CopyDVD\smoothdeinterlacer\SmoothDeinterlacer.dll")
avisource("D:\CAPTURE\palcap.avi")
Trim(40285,59711)
crop(4,12,688,552)
SeparateFields()
odd=SelectOdd.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
evn=SelectEven.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
Interleave(evn,odd)
Weave()
AddBorders(16,12,16,12)
SmoothDeinterlace(tff=true, doublerate=true)
LanczosResize(720,480)
ChangeFPS(59.94)
SeparateFields()
SelectEvery(4,1,2)
Weave()
ConvertToYUY2(interlaced=true)

This capture was done at 696x576 which is the resolution my el wierdo capture card (an AverTV Stereo) needs to captue at a correct aspect ratio. Capture was done using PICVideo MJPEG top field first.

Please note I used AviSynth 2.54 and the AviSynth 2.5x version of SmoothDeinterlacer as well as Convolution3D.

Also I encoded with CCE but if you will be using TMPGEnc then you should change the last line from ConvertToYUY2(interlaced=true) to ConvertToRGB(interlaced=true)

Worked for me!

- John "FulciLives" Coleman

[erratic]

SmoothDeinterlace(tff=true, doublerate=true)

I didn't know SmoothDeinterlace had a field order option (I don't use it anymore). In that case SmoothDeinterlace(tff=true, doublerate=true) is the same as AssumeTFF().SmoothDeinterlace(doublerate=true) of course.

But SmoothDeinterlace definitely creates a double frame (easy to verify). Here's what scharfis_brain (biggest avisynth expert I know) wrote in Doom9's forum: it isn't in phase with all other bobbers out there (it is one frame behind/future, sorry can't tell you now).
Removing that double frame is better, even though ignoring it it doesn't cause jerkiness, so if you guys are happy with the quality you're getting, that's fine by me. As scharfis_brain writes in this thread, his kernelbob function (for KernelDeint by Donald Graft) is better than SmoothDeinterlace(doublerate=true). I've dumped SmoothDeinterlacer in favor of kernelbob. But I guess it's too complicated for some people, so never mind.

Also ConvertFPS is slightly better than ChangeFPS to avoid jerkiness (but it's probably slower as well). Once again I'm quoting scharfis_brain: changefps only duplicates/drops frames to change the framerate, while convertfps mixes the nearby frames together to achieve a better motion-feeling.

[NamPla]

Success! Yes, my mistake was definitely not specifying the "Assume TFF"!

I tried both Erratic's and FulciLives's scripts...only had success with FulciLives, which puzzled me a moment because I thought they were both doing basically the same things. Then I realized I had previously writted:

ConvertToRGB()

...but it should have been:

ConvertToRGB(interlaced=true)

That was my mistake (not Erratics).

Anyhow it works great, quality is great!

Thanks so much!

(Now I have to look at this "KernelBob" etc, he!) :P

[NamPla]

Hi again!

I'm getting fabulous quality with the PAL-to-NTSC conversions, combined with the Convolution3D avisynth script is really great!

But I've just struck an "Access Violation" error!!! Help!

I notice if theres a slight glitch in the capture avi (ie picture breaks up momentarily) then avisynth rejects it. That means from that point on, the encoding is black screen with error message - even if just a moment glitch in the middle of the avi! Ugh, lots of time wasted...

Unfortunately this is a PAL capture that I need to convert to NTSC (with avisynth). So I NEED to use avisynth! But avisynth is rejecting the avi file!!

I'm not sure what to do? Can anybody pleaes help? I'm wondering if I can do something with my script that can somehow ignore the "glitch"???

Thank you!

[FulciLives]

@NamPla

I've never encountered the problem you are talking about.

Can you explain these "glitches in your capture AVI" in perhaps more detail?

- John "FulciLives" Coleman

[NamPla]

Hi FulciLives. What I mean is, the VHS is a bad source I guess. For a second or two there is some breakup of the picture, scrolling horizontal noise lines, like tracking gone wrong for a second. But then it restabilizes and is OK. It's definately the tape, not my VCR or heads or anything.

AND... I've duplicated this problem with ANOTHER tape, that ALSO has a "glitch" like this!. When I feed the avisynth script into TMPGEnc, when it comes to the glitch, the encoding black-screens with the red "Access Violation" warning message, right to the end of the encode.

So it happens with 2 of my tapes now. Avisynth isn't liking tracking noise on the avi capture. Maybe it's working so FAST, that it can't handle a sudden random noise, even for a second?

Hope I'm explaining this to you OK?

Anyhow, I figure I may have to resort to my old VirtualDub frameserving for these problematic captures...EXCEPT, I need to convert them PAL-to-NTSC!?!?!! Arrrgghh...

[FulciLives]

Again I'm not sure what to tell you as I never encountered this problem myself.

If you can try to use another encoder such as Cinema Craft Encoder or Mainconcept MPEG encoder then try that.

Otherwise I don't know what to tell you ... sorry!

- John "FulciLives" Coleman