AviSynth 2.5x and Convolution3D guide for AVI captures

[NamPla]

That's Ok John. I read around a bit & learned "Access Violation" occurs if run out of RAM (heavy-duty processing)...or corrupted avi source.

I simply trimmed out the offending one second of glitch and the encoding was OK then.

Maybe you haven't struck this problem because your sources are good? So are mine generally, but this was music footage, on an old VHS... Avisynth doesn't like corrupt video...!!!

BTW I'm hoping to migrate to CCE soon...as soon as I can spare the cash!!!

CHEERS!

[NamPla]

Get this, I have my virtual memory set to "custom" mode (2x RAM = 1024MB). So I changed it back to0 "system managed" mode as an experiment, and now there's NO MORE "access violation" alerts from avisynth! Now it accepts all my avis, glitches & all!

Just thought I'd report it here, in case any body else finds it interestings.

Thanks.

[FulciLives]

Get this, I have my virtual memory set to "custom" mode (2x RAM = 1024MB). So I changed it back to0 "system managed" mode as an experiment, and now there's NO MORE "access violation" alerts from avisynth! Now it accepts all my avis, glitches & all!

Just thought I'd report it here, in case any body else finds it interestings.

Thanks.

I've been away ... having computer problems here after my boot HDD crashed due to a hardware failure (it was about 4 years old).

Anyways I am glad that you found a solution to your problem

- John "FulciLives" Coleman

[NamPla]

I did an experiment & installed the ReadAVS VFAPI plug-in into TMPGEnc. That way I can chop out the "ConvertToRGB" line & everything stays in YUY2 colour-space. (Just put "ConvertToYUY2(interlaced=true)"!!)

I noticed encoding time "slightly" sped up (I think) and skin tones looked more natural!

[Bizuser]

I followed the guide, and got great results (thanks!) with one change:

Here's the original script:

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() 
DoubleWeave.SelectOdd()

When I used this, frameserving into TMPGEnc, I got 'jerky' results. In TMPG, on the advanced tab, I had 'Bottom Field First' selected - this was the default (I believe TMPG remembers the last used setting, and I had been processing some DV AVIs 'natively' (without avisynth) so BFF was appropriate).

When I switched to TFF, the problem went away, but I am guessing that it's better to reduce the number of steps in the Avisynth script file, so I commented out the last line (DoubleWeave) - is that correct? Your instructions say to comment that out if your SOURCE is TFF also.

Another question .... which of the presets would you recommend for a good quality capture from cable using a canopus advc 100? Not sure how this relates to 'low quality movie', etc.

And one final question ... is there a reason I can't set Bitrate in TMPG above 8000? Shouldn't I be able to go to 9000?

Thanks again for the guide!

[jimmalenko]

@Bizuser:
Just wanted to let you know that Fulci hasn't been sighted at the forum for a while now. I can answer your questions, so I will !

When I used this, frameserving into TMPGEnc, I got 'jerky' results. In TMPG, on the advanced tab, I had 'Bottom Field First' selected - this was the default (I believe TMPG remembers the last used setting, and I had been processing some DV AVIs 'natively' (without avisynth) so BFF was appropriate).

When I switched to TFF, the problem went away, but I am guessing that it's better to reduce the number of steps in the Avisynth script file, so I commented out the last line (DoubleWeave) - is that correct? Your instructions say to comment that out if your SOURCE is TFF also.

I had that exact same issue. I was capturing from an ADVC-100 using the Canopus codec and had TMPGEnc set to BFF. My take on it is that the original guide is worded wrong and like you, I commented out that last line and this removed the "jerkyness".

Another question .... which of the presets would you recommend for a good quality capture from cable using a canopus advc 100? Not sure how this relates to 'low quality movie', etc.

I have found that the default settings provided in this guide provide the best quality/time trade off. If your source is good then this just puts the polish on it. If your source is bad, it cleans it up a little - garbage in=garbage out, right ? I also found that when I jacked up the settings to whatever the worst one is, the colors were effected and everything seemed to have too much red in it.

And one final question ... is there a reason I can't set Bitrate in TMPG above 8000? Shouldn't I be able to go to 9000?

You need to get out of the wizard. It only lets you go to 8000kbps, whereas the "manual" way lets you go above 8000kbps.

[NamPla]

Yes I'm worried I scared FulciLives away!!! :P

I hope he's alright, I loved his input into this forum.

Yes his Convo3D settings are great for TV or DVD etc... VHS works better with more severe settings. But of course it all depends on your equipment & cables. I've used the C3D "anime" settings for video & it's worked great too. You must experiment. I find "1" looks sharper than "0" at the start of the C3D parameters.

And also, downsize to 352x576/480 - despite some of the shit on this forum this is the best method (depending on your capture card, OK). No "8000-9000" bit, you only need 5000-bit max (2000 min/100%qual/VBR_CQ) at 352x576/480 and then you can include the PCM wav audio & more stuff etc...And you don't lose any quality anyway...

BUT... I am of course interested to hear other takes of this...

[Bizuser]

This guide (the subject of this forum) is almost identical to a guide over on doom9 (http://www.doom9.org/dv/guide.html). The one difference is that over there, they have this to say about NTSC. ...

NTSC issues

As NTSC DV uses the 4:1:1 colorspace, there is an issue with the conversion to the 4:2:0 colorspace used in DVD MPEG-2. To address this issue use trbarry's filter ReInterpolate411().

To get improved color conversion, simply add a
ReInterpolate411()
command below the AviSource or DirectShowSource command, respectively.

Do you guys have any opinion on this? I tried it, and it looks good, but I haven't been able to figure out what it's really doing!

Regarding downsizing to 352x576/480 ... I tried that once a long time ago and got poor results. Maybe I CAPTURED at that res with a capture card, though. You are saying capture at full res, then downsize. I can see this applying to VHS captures, but what about high quality Cable material? I'm working on a 'Nova' show right now, and the material is quite good. I know I can test and see for myself, but I'm testing 50,000 other things too and this was not a variable I was considering! Right now, I'm happy to devote one blank DVD to the one hour show, but for the future, the concept is useful.

Regarding 8,000 max bitrate ... I just can't seem to get past this ... I'm not using the wizard. The field is open (I can reduce it, but can't exceed 8k). My show is only 1 hr and I have room to spare (not that this matters to TMPG of course!). Anyway, I'll figure that out. IF I downsize the material to 352x, then I can imagine I can use a much lower bitrate, but if I stick to 720x, then I do believe I need the 8k or more. The quality/smoothness of panning shots was definitely improved by going from 7k to 8k, and I want to try going even higher.

[BSR]

This guide (the subject of this forum) is almost identical to a guide over on doom9 (http://www.doom9.org/dv/guide.html). The one difference is that over there, they have this to say about NTSC. ...

NTSC issues

As NTSC DV uses the 4:1:1 colorspace, there is an issue with the conversion to the 4:2:0 colorspace used in DVD MPEG-2. To address this issue use trbarry's filter ReInterpolate411().

To get improved color conversion, simply add a
ReInterpolate411()
command below the AviSource or DirectShowSource command, respectively.

Do you guys have any opinion on this? I tried it, and it looks good, but I haven't been able to figure out what it's really doing!

I suggest using it, I have seen it posted many of times about people using this. I haven't noticed exaclty what it does, but then again I didn't really try to look for differences either. I know Fulci did a lot of research over at Doom9 to write this guide, not sure why he left that out.

Regarding downsizing to 352x576/480 ... I tried that once a long time ago and got poor results. Maybe I CAPTURED at that res with a capture card, though. You are saying capture at full res, then downsize. I can see this applying to VHS captures, but what about high quality Cable material? I'm working on a 'Nova' show right now, and the material is quite good. I know I can test and see for myself, but I'm testing 50,000 other things too and this was not a variable I was considering! Right now, I'm happy to devote one blank DVD to the one hour show, but for the future, the concept is useful.

To each his own. For material I care about, I use full res. For material not as important and I need to put more info on a DVD, then half res.
I see a noticable difference with half res, even with VHS captures.

Regarding 8,000 max bitrate ... I just can't seem to get past this ... I'm not using the wizard. The field is open (I can reduce it, but can't exceed 8k). My show is only 1 hr and I have room to spare (not that this matters to TMPG of course!). Anyway, I'll figure that out. IF I downsize the material to 352x, then I can imagine I can use a much lower bitrate, but if I stick to 720x, then I do believe I need the 8k or more. The quality/smoothness of panning shots was definitely improved by going from 7k to 8k, and I want to try going even higher.

Wonder if TMPGEnc might limit you to 8000 if your using PCM audio because of the extra bandwidth that PCM takes up. I'll have to check on this myself.

We'll have to start a have you seen Fulci thread. Hopefully he's just taking a break and will return.

[Bizuser]

One thing I've noticed is that either the Convolution3D or the ReInterpolate411() (or both) are extremely slow! With DC precision 10, Motion Search 'High' (not 'Highest'), it's taking 14 hours for TMPG to encode a 2 hr show! This is on my 2.4 GHz P4 box. Typical encode times up until now have been in the 4-6 hour range, with motion search set to 'Highest'!

Oh well, maybe it's worth it! I'll have to try this on my work PC, which is 3.2 GHz and supposedly HT enabled.

As for bitrate over 8k, even when encoding with Mpeg 1 layer 2 audio (334 kbps), I could not go above 8k. However, I've read that that can lead to trouble anyway with some media and/or some players, so maybe it's best to leave this alone. At other times, though, I've been able to set bitrate over 8k ... so there is SOME setting somewhere in TMPG that is causing this. Not a big deal - I'm obviously doing something wrong!

[BSR]

Convolution 3D is extremely slow.
I use Motion estimate Search, that helps speed up the process and gives great results. You'll see a lot of guides recommend this.
If your using 2pass VBR then that time doesn't sound that out of wack with convolution 3D, but does seem on the high side.
Not sure why you can't go above 8K bitrate, tested mine and it works.

[newbie04]

Hey all,
umm this might be a dumb question but the very last part of the guide:

"Anyways at this point you simply load your AviSynth AVS script into TMPGEnc (or CCE) just as if you were loading the original AVI file and go ahead and do your encoding. "

I have DVD2SVCD and it came with CCE. I looked around in the DVD2SVCD tabs but found no way to like make a custom script? As for CCE by itself, I don't know how you load avs scripts in it either! Please tell me how exactly do I load this script?? Thanks!

[erratic]

I'm not familiar with DVD2SVCD so I can't help you with that, but you can load avs scripts in CCE just like you load ordinary avi files. Drag&drop the avs script onto the main CCE window or select File -> Open -> File.

[jyn999]

In DVD2SVCD, click and open the Frameserver tab. Type the name of your AviSynth file in the AviSynth setup box.

Just remembered a tip. Select the Edit as part of Video encoding radio button. CCE encoding will stop and allow you to edit your AviSynth script if you want to. Click the OK button and CCE will continue processing the AviSynth script complete with any changes.

[newbie04]

In DVD2SVCD, click and open the Frameserver tab. Type the name of your AviSynth file in the AviSynth setup box.

I wasn't able to do that (you can't type anything), it has presets, but I read the Q & A, and it tells about making your own preset.

Just remembered a tip. Select the Edit as part of Video encoding radio button. CCE encoding will stop and allow you to edit your AviSynth script if you want to. Click the OK button and CCE will continue processing the AviSynth script complete with any changes.

Thanks for this! When I did that a box popped up like this:


I don't know which method is better, but at least I know what to do now. Thanks so much for the help!

[newbie04]

This guide (the subject of this forum) is almost identical to a guide over on doom9 (http://www.doom9.org/dv/guide.html). The one difference is that over there, they have this to say about NTSC. ...

NTSC issues

As NTSC DV uses the 4:1:1 colorspace, there is an issue with the conversion to the 4:2:0 colorspace used in DVD MPEG-2. To address this issue use trbarry's filter ReInterpolate411().

To get improved color conversion, simply add a
ReInterpolate411()
command below the AviSource or DirectShowSource command, respectively.

Do you guys have any opinion on this? I tried it, and it looks good, but I haven't been able to figure out what it's really doing!

I suggest using it, I have seen it posted many of times about people using this. I haven't noticed exaclty what it does, but then again I didn't really try to look for differences either. I know Fulci did a lot of research over at Doom9 to write this guide, not sure why he left that out.

I found this description:
ReInterpolate411 - 2003/07/31 V 0.1.1

This is a fast simple filter to correct the improper 4:1:1 => 4:2:2 conversion that seems to occur with some DV/4:1:1 codes when used with Avisynth. It assumes the odd chroma pixels are duplicates and discards them replacing them with the average of the two horizontally adjacent even chroma pixels. It doesn't matter whether the clip is interlaced though it must be in YUY2 format for Avsynth 2.5. There are no parms, and currently no readme file.

Though that is way too technical for me to understand.

[Wilbert]

4:1:1 layout

YC1 Y Y Y YC3 Y Y Y ...
YC1 Y Y Y YC3 Y Y Y ...
YC1 Y Y Y YC3 Y Y Y ...
...

the mainconcept codec duplicates the missing chroma (when converting to YUY2):

YC1 Y YC2x Y YC3 Y YC4x Y ...
YC1 Y YC2x Y YC3 Y YC4x Y ...
YC1 Y YC2x Y YC3 Y YC4x Y ...
...

with C2x = C1, C4x = C3.

ReInterpolate411 replaces the duplicates with interpolations:

C2x = (C1+C3)/2, C4x = (C3+C5)/2

Note, you only need this when using the MainConcept codec.

If you look around at doom9's forum (dv section). You will also find ReInterpolate420 correcting a similar problem for pal (here chroma is horinzontally dupicated instead of interpolated).

[dbarndt]

Hi,

I know it's been a while since anything was posted here but I just wanted to add a quick observation.

Thanks to this note, I am editing away with VirtualDubMod and AVISynth, and the results have been great! But I found one thing - capturing NTSC with my ADVC-50 and WinDV, the DV came out TFF, not BFF (I was seeing those horizontal "jitters"). Since I'm doing pretty much what FulciLives' original script is doing, I simply commented out the line in the original script:

DoubleWeave.SelectOdd()

...and it seemed to work.

Here's my setup and the script I'm using. Hope it can help someone else.

ADVC-50, WinDV
VirtualDubMod, AVISynth
TMPGEnc Plus 2.5

#
# Load plugins
#
# - convolution 3D filter
LoadPlugin("Convolution3d.dll")
#
# Get AVI DV source (Canopus ADVC-50 => WinDV, produces RGB24 bit colorspace)
src1 = AVISource("")
clip1 = Trim(src1, , ).FadeIn(5).FadeOut(5)
.
.
clipN = Trim(src1, , ).FadeIn(5).FadeOut(5)
#
# Create blank "pad" for use between scenes
#
pad = BlankClip(length=30, width=720, height=480, pixel_type="RGB24", fps=30000, fps_denominator=1001, audio_rate=48000, stereo=true, color=$000000)
#
# Splice everything together
#
UnalignedSplice(pad, clip1, pad, ... pad, clipN, pad)
#
# Crop if desired (left, top, right(neg), bottom(neg))
#
#Crop(0,0,0,-52)
#
# Convert to YUY2 colorspace for convolution 3D filter
#
ConvertToYUY2(interlaced=true)
#
# Apply convolution 3D filter to interlaced video
#
SeparateFields()
# - Movie High Quality (ex. good DVD source)
#odd=SelectOdd.Convolution3D (0, 3, 4, 3, 4, 2.8, 0)
#evn=SelectEven.Convolution3D (0, 3, 4, 3, 4, 2.8, 0)
# - Movie Low Quality (ex. noisy DVD source or most captures)
odd=SelectOdd.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
evn=SelectEven.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
# - VHS capture Bad Quality (ex. some captures)
#odd = SelectOdd.Convolution3D(0, 32, 128, 16, 64, 10, 0)
#evn = SelectEven.Convolution3D(0, 32, 128, 16, 64, 10, 0)
Interleave(evn, odd)
Weave()
#
# Deinterlace only if converting PAL -> NTSC
#
#SmoothDeinterlace(tff=true, doublerate=true)	# top field first
#SmoothDeinterlace(tff=false, doublerate=true)	# bottom field first
#Trim(1,0)
#
# Use only if NOT deinterlacing and if BFF (bottom field first);
# Canopus ADVC-50 => WinDV capture seems to be TFF (top field first)
#
#DoubleWeave.SelectOdd()
#
# If cropped above, add borders (left, top, right, bottom) where cropped
#
#AddBorders(0,26,0,26)
#
# Resize if desired
#
#LanczosResize(720,480) # if downsizing
#Lanczos4Resize(720,480) # if upsizing (not recommended; seems to cause combing effects)
#
# Conversion for TMPGEnc not required if it has the ReadAVS plugin
#
#ConvertToRGB24(interlaced=true)

Thanks again for a GREAT guide and a GREAT forum!

Dave

[FulciLives]

#
# Load plugins
#
# - convolution 3D filter
LoadPlugin("Convolution3d.dll")
#
# Get AVI DV source (Canopus ADVC-50 => WinDV, produces RGB24 bit colorspace)
src1 = AVISource("")
clip1 = Trim(src1, , ).FadeIn(5).FadeOut(5)
.
.
clipN = Trim(src1, , ).FadeIn(5).FadeOut(5)
#
# Create blank "pad" for use between scenes
#
pad = BlankClip(length=30, width=720, height=480, pixel_type="RGB24", fps=30000, fps_denominator=1001, audio_rate=48000, stereo=true, color=$000000)
#
# Splice everything together
#
UnalignedSplice(pad, clip1, pad, ... pad, clipN, pad)
#
# Crop if desired (left, top, right(neg), bottom(neg))
#
#Crop(0,0,0,-52)
#
# Convert to YUY2 colorspace for convolution 3D filter
#
ConvertToYUY2(interlaced=true)
#
# Apply convolution 3D filter to interlaced video
#
SeparateFields()
# - Movie High Quality (ex. good DVD source)
#odd=SelectOdd.Convolution3D (0, 3, 4, 3, 4, 2.8, 0)
#evn=SelectEven.Convolution3D (0, 3, 4, 3, 4, 2.8, 0)
# - Movie Low Quality (ex. noisy DVD source or most captures)
odd=SelectOdd.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
evn=SelectEven.Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
# - VHS capture Bad Quality (ex. some captures)
#odd = SelectOdd.Convolution3D(0, 32, 128, 16, 64, 10, 0)
#evn = SelectEven.Convolution3D(0, 32, 128, 16, 64, 10, 0)
Interleave(evn, odd)
Weave()
#
# Deinterlace only if converting PAL -> NTSC
#
#SmoothDeinterlace(tff=true, doublerate=true)	# top field first
#SmoothDeinterlace(tff=false, doublerate=true)	# bottom field first
#Trim(1,0)
#
# Use only if NOT deinterlacing and if BFF (bottom field first);
# Canopus ADVC-50 => WinDV capture seems to be TFF (top field first)
#
#DoubleWeave.SelectOdd()
#
# If cropped above, add borders (left, top, right, bottom) where cropped
#
#AddBorders(0,26,0,26)
#
# Resize if desired
#
#LanczosResize(720,480) # if downsizing
#Lanczos4Resize(720,480) # if upsizing (not recommended; seems to cause combing effects)
#
# Conversion for TMPGEnc not required if it has the ReadAVS plugin
#
#ConvertToRGB24(interlaced=true)

Dave,

Just curious where you got this sample script from? It has some interesting things in it such as the BLANK CLIP part.

- John "FulciLives" Coleman

[dbarndt]

Hi John,

Sorry for the long delay replying - I just happened to check back here and saw your note.

The script I'm using has simply evolved from the ones given here as examples and from the "manual" at the AVISynth web site (ex. that's where I found BlankClip() to put in "space" between clips).

I've found that if you use FadeOut() with values higher than 5-10 frames the audio sometimes "spikes" (for lack of a better word) on the way down. I wonder if anyone else has seen this.

I use this script for EVERYTHING mainly because of the Convolution3D filter. I can't thank you enough for getting me started off on the right foot...!

Dave

[FulciLives]

I use this script for EVERYTHING mainly because of the Convolution3D filter. I can't thank you enough for getting me started off on the right foot...!

I'm glad you found it helpfull but really I got the script, more or less, from the
DOOM9 website ... but a lot of people here seem to think AviSynth is "scary" so I am always happy to find out that my guide has been helpfull

- John "FulciLives" Coleman

[SatStorm]

Your guide is helpfull..
And Avisynth is Scary!

[Wile_E]

DoubleWeave.SelectOdd()

This gave me stuttering effect on my bottom field DV. TmpGenc kept telling me I had Top Field video. But my original DV file is bottom. So I had to change it to the following to get the filter to work right:
DoubleWeave.SelectEven()

[jimmalenko]

DoubleWeave.SelectOdd()

This gave me stuttering effect on my bottom field DV. TmpGenc kept telling me I had Top Field video. But my original DV file is bottom. So I had to change it to the following to get the filter to work right:
DoubleWeave.SelectEven()

I had the same issue, but I simple just deleted the line altogether to get my working properly.

[Sillyname]

Will I benefit from using a Dual Core Pentium during this process?

[FulciLives]

Will I benefit from using a Dual Core Pentium during this process?

I have no idea but more CPU power generally means faster encoding time.

- John "FulciLives" Coleman

[Sillyname]

Will I benefit from using a Dual Core Pentium during this process?

I have no idea but more CPU power generally means faster encoding time.

- John "FulciLives" Coleman

But does VirtualDub utilize dual processors in it's hand off to the encoder I choose?

[FulciLives]

Will I benefit from using a Dual Core Pentium during this process?

I have no idea but more CPU power generally means faster encoding time.

- John "FulciLives" Coleman

But does VirtualDub utilize dual processors in it's hand off to the encoder I choose?

You are not using VirtualDubMod to frameserve. You are only using it to edit and get your edit points for the AviSynth script. You load the AviSynth script directly into your MPEG encoder.

- John "FulciLives" Coleman

[Sillyname]

Will I benefit from using a Dual Core Pentium during this process?

I have no idea but more CPU power generally means faster encoding time.

- John "FulciLives" Coleman

But does VirtualDub utilize dual processors in it's hand off to the encoder I choose?

You are not using VirtualDubMod to frameserve. You are only using it to edit and get your edit points for the AviSynth script. You load the AviSynth script directly into your MPEG encoder.

- John "FulciLives" Coleman

So then this is my problem. I will be trimming in an editor ahead of time. I need it to be DV on my final output, not mpeg2. Will a DV encoder take *.avs files?

[Sillyname]

I get an error code 3 when I try to bring the AVS into virtualdubmod.

I tried just opening one of my DV files in virtualdubmod and it says something about the compressor not being there. B.S. I can play DV files in every player we have on the machine BUT virtualdubmod. I started to try to install the Panasonic DV codec but realized how ridiculous this was. What can I do to make virtualdubmod read a DV Type 2 NTSC avi? Can I point it to a DV codec I already have existing on the computer?

I really intend to use this for something rather important. I don't have much time to patch holes in freeware but I am going on the advice of people here, that this is something special. Please help. I'm sorry I come off as being inept.