TMPGEnc for HD to SD conversion?

[Canon GL-2 Guy]

I read in another major video forum that this tool may do a appreciably better job when downscaling using the Lanczos3 algorithm/ filter especially when compared to major NLE's such as Prem Pro and Vegas both of which utilize Main Concepts encoder which yields poor results.

Is there any truth to that statement? Which flavor of TMPGEnc was being referred too? The downside is that it costs $100.00.

Any thoughts on that claim?

It was also stated that Virtuadub which is FREE uses the same filter to yield good results, however you have 1st to feed it a high Q .avi file such as Cineform/Laragith or say a 100 mbps MPEG-2 I-Frame 4:2:2 obtained via Adobe Media Encoder.

Bottomline will VD work without the use of Avisynth and other tools which are demonstarted in Bellunes video?

Is there a satisifying solution that falls between between "pristine" and awful?

As far as the downscaling from HDV 16:9 to SD 16:9 widescreen goes:

1) Has CS 5.5 improved it's downscaling capability & quality over say version CS 3.2?
2) Have any of you tried Cineforms Neo which is rather pricy for this specific task?

Also, what can one do to sharpen a SD file that was just downscaled from HD? Is the sharpening applied before after downsizing?

Thanks in advance.

[jagabo]

If your video is very sharp to start with a sharpening filter like Lanczos3 will result in oversharpening halos and increased noise. A Bilinear (Precise Bilinear in VirtualDub) will give more natural results in that case. If your video isn't so sharp and doesn't have much noise (or you don't care about increasing noise a bit) a sharpening filter like Bicubic, Lanzsoz3, Spline36, etc. will look good.

For interlaced video, the best method is to convert to double frame rate progressive with a good smart bob filter, then downsize, and interlace. Like:

WhateverSource() # 30i source
QTGMC() # convert to 60p
BilinearResize(720,480) # resize for DVD
SeparateFields() #convert to fields
SelectEvery(4,0,3) # select top and bottom fields
Weave() # weave them into frames

The cheap trick of separating fields, resizing them, then weaving them back into frames, works ok for not-so-sharp camcorder video.

[poisondeathray]

I read in another major video forum that this tool may do a appreciably better job when downscaling using the Lanczos3 algorithm/ filter especially when compared to major NLE's such as Prem Pro and Vegas both of which utilize Main Concepts encoder which yields poor results.

It's not Mainconcept's fault. It doesn't do any scaling, that's done by PP. The scaling is applied before the encoding.

Is there any truth to that statement? Which flavor of TMPGEnc was being referred too? The downside is that it costs $100.00.

Any thoughts on that claim?

You should ask in that forum, how can someone here answer what version was being referred to over there ?

I think there is a free trial available, so give it a shot

It was also stated that Virtuadub which is FREE uses the same filter to yield good results, however you have 1st to feed it a high Q .avi file such as Cineform/Laragith or say a 100 mbps MPEG-2 I-Frame 4:2:2 obtained via Adobe Media Encoder.

Bottomline will VD work without the use of Avisynth and other tools which are demonstarted in Bellunes video?

Note the "HD2SD" script is different than the vdub lanczos interlace resize method.

You don't have to use avisynth for vdub, but the "interlaced scaling" checkbox has to be checked. You might want to import into vdub using avisynth, because it means you skip the first step of encoding to the digital intermediate or using huge filesize imports. Vdub natively doesn't handle interlaced YV12 ideally. It upsamples as progressive and you get chroma artifacts so if you don't use avisynth to upsample , or upsample using cineform or MPEG 4:2:2 before importing into vdub, you will get problems

With avisynth you can frameserve in and out without using any huge intermediate files . But if you frameserve, you don't even need vdub: You can do it all in avisynth and import .avs into premiere directly

The benefit of using a cineform intermediate is that editing in HD timeline will be smoother, with a very slight quality loss tradeoff . Frameserver adds latency overhead on top of the original files, so HD files won't be very smooth to edit (HDV not so bad, but AVCHD is a pig)

Is there a satisifying solution that falls between between "pristine" and awful?

You should be aware there is no such thing as "pristine" in interlaced scaling. There are always tradeoffs and artifacts. Interlace is inherently flawed to begin with.

"Sharpness" can be bad for SD interlace because it accentuates the perception of aliasing and line twittering.
http://en.wikipedia.org/wiki/Interlaced_video#Interline_twitter

This is why many algorithms apply a LOW PASS and make the image less crisp. On fine lines and patterns like striped shirts, you will almost always get line buzzing artifacts without low passing. But if you low pass the entire composition, it will be blurry. Some people apply filters selectively as to only degrade those sections that need it. Some viewers are very sensitive to this, some less so

As far as the downscaling from HDV 16:9 to SD 16:9 widescreen goes:

1) Has CS 5.5 improved it's downscaling capability & quality over say version CS 3.2?
2) Have any of you tried Cineforms Neo which is rather pricy for this specific task?

Also, what can one do to sharpen a SD file that was just downscaled from HD? Is the sharpening applied before after downsizing?

Yes it has improved , if you have CUDA enabled and MRQ enabled it will use Lanczos 2 (not 3) with low pass bicubic. Lanczos 3 is a sharper algorithm but has the potential to create more artifacts (ringing) as jagabo mentioned . Too sharp is not a good thing, especiallly with interlaced video

Yes MRQ, No Cuda = Bicubic scaling
No MRQ, No Cuda = Bilinear with gaussian low pass



When applying sharpening , you usually do it after downscaling, but it has to be an interlace aware "sharpen" filter.
Not only is it faster (Fewer pixels to process), it's more accurate

[Canon GL-2 Guy]

Gents,

You've given me very good answers and things to think about!

I will respond with some pointed questions based on the above soon.

If anyone else wants to pitch in about their HD to SD experience with TMPGEnc that would be a bonus.

Thanks

[Canon GL-2 Guy]

I'm back.

Source footage is HDV 1080 60i. Objective is SD widescreen.

I do not have Cuda or MRQ enabled yet. That will require installing and finding a power supply that will fit my HP case which is on the agenda.

Jagabo, please would you mind breaking down your instruction step by step and how it would be achieved so that a novice to the process could follow it?

"For interlaced video, the best method is to convert to double frame rate progressive with a good smart bob filter, then downsize, and interlace". How would I achieve the conversion?

1) Of the 3 methods listed in my first post which could be used to create the source .avi, which one offers the most pros? I have a WD 2 TB drive on order to handle the anticipated read that huge immediatory files.

2) I'm not sure I installed all of the programs in Bellunes video tutorial right, know nothing about avisynth et. al. and on the surface seems involved and complicated. Would like to see Virtualdub do the downscaling on it's own if feasible. Any other must have settings or considerations using VB? Alternatively where would I find a AVIsynth script file/string that I could just feed in?

My master footage varies in sharpness dependent on how good or poor the lighting is in my shooting environment which is the interior of a church. Therefore the results vary. What filter would you recommend as the best "all around" starter?

What really bothers me is that when a pastor wears a sharp, nicely iron dress shirt the edges/straight lines on playback is an absolute mess which is especially accentuated in SD. Talk about a bad case of the jaggies. Any ideas there?

If I use Lagarith or generally, what colorspace should I be working in keeping in mind that my only NLE is several versions of Premiere Pro.

Thanks

[poisondeathray]

Personally, I would use an HD timeline to edit, render out a HD master . This gives you more options later on should you decide to make a blu-ray or HD format for web. Also editing native HDV is quite low in terms of system resources, and editing should be pretty snappy

From that HD master, I would apply the whaterver variation of avisynth script you choose to use . The colormatrix must be converted to Rec601 for SD footage. If you use HD2SD() , it does all this for you including the Rec709=>Rec601 conversion.

HD2SD is very easy to use . If you rendered out a cineform "master" AVI, the script would be . BTW, cineform has a free version (just for encoding and decoding), ever since gopro bought them out. So you can still use cineform, but you don't have HDLink, Firstlight or other accessory functions

AVISource("cineform.avi")
HD2SD()

That's it. 2 text lines and you feed that into any encoder that accepts avs scripts e.g. hcenc , cce, even AME accepts .avs scripts with the avs import plugin if you wanted to use Adobe's mainconcept encoder



I used to use TGMC (precursor to QTGMC) and almost that exact same script that jagabo gave above. The speed IMO isn't worth it. It a very tiny difference in quality once it's resized to SD and reinterlaced , even compared to simple bob() + rest of script. You decide if the speed tradeoff is worth it for you. You can use a faster preset for QTGMC if you want. Also note the default QTGMC settings are very sharp (oversharp IMO, even for progressive footage). I usually turn it way down

For the striped shirts and problematic footage areas, I use a script called IResize()
http://forum.doom9.org/showthread.php?p=1185790#post1185790

It produces a lot softer results, but virtually zero aliasing/jaggies. You should see the comparisons on the "torture" test clip; I can't find the thread right now... but I'll try to dig it up. Amazing. Anyways I would rotoscope or mask out sections that require it, and use the generic script for the other sections that you want to keep sharp

[Canon GL-2 Guy]

I have had Neoscene for some time waiting in the wings just in case.

Recently I finally had a firewire card installed, so in addition to exporting the completed master project onto tape for archival purposes, I started to export the entire timeline via CS 5.5 to obtain these types of files:

1) Cineform YUV 4:2:2 @ high quality
2) MPEG-2 I -Frame 100 Mbps/.m2t transport stream
3) Take file from number 2 and then have Neoscene convert it to Cineform.

Thus far, I have not got CS 5.5 to export the master timeline to Lagarith which makes no sense to me.
Can't get the settings on the video tab to be 1440 by 1080.

What is HD2SD anyways: a script, or a program/filter residing within AVIsnyths plug-in folder?

[Canon GL-2 Guy]

Would you be willing to read my seperate post on CS 5.5 Lagarith and AME please?

All this ties into together.

I'll just toss this next question out there!

Is there a way to convert 60i footage to 60p so that's it's "easier" to work with?

I think my next ENG camcorder will be 60p.

[Canon GL-2 Guy]

PS) I can't afford or justify $500 for Main Concepts Reference encoder, although that sure would be nice to have as a resource.

[poisondeathray]

Would you be willing to read my seperate post on CS 5.5 Lagarith and AME please?

All this ties into together.

I'll just toss this next question out there!

Is there a way to convert 60i footage to 60p so that's it's "easier" to work with?

I think my next ENG camcorder will be 60p.

I'll answer that in a second

60i => 60p is called "bob deinterlacing" or "double rate deinterlacing"

The QTGMC line in jagabo's script is an example of an excellent motion compensated bob deinterlacer

At full resolution, the differences can be very large between deinterlacers. But once it's be deinterlaced, resized to SD and re-interlaced, the big differences are much smaller

[jagabo]

fast, looks ok with material that's not too sharp:

SeparateFields()
BilinearResize(720,240)
Weave()

pretty fast, has trouble with near parallel lines and sharp edges

Yadif(mode=1)
BilinearResize(720,480)
SeparateFields()
SelectEvery(4,0,3)
Weave()

faster variation of QTGMC

QTGMC(preset="Very fast")
BilinearResize(720,480)
SeparateFields()
SelectEvery(4,0,3)
Weave()

[Canon GL-2 Guy]

I now feel "dumb" however somehow my alias/shorcut for avisynth dissapeared or perhaps got deleted.

Don't want to uninstall and then have to reinstall the whole Bellune process again!

Revouninstaller and others claim AVS & the resy of the bunch is still installed, however this is no application icon or shortcut to be found.

I want to try your scripts. Is there still someway to launch AVS or is it supposed to be launchable?

[jagabo]

AviSynth works in the background. You create a text based script with commands like those listed above. Then you open that script with an editor that supports AviSynth scripts. When you installed AviSynth it should have associated Notepad with the .AVS extension. If not, you can easily do it yourself. Start with something simple:

http://avisynth.org/mediawiki/First_script
http://avisynth.org/mediawiki/Getting_started

[Canon GL-2 Guy]

I rarely if ever use Notepad for anything however I guess it's time to change.

How to I associate this app with AVS as a default?

[Canon GL-2 Guy]

I've used Word and Excel for so long I forgot it existed.

[jagabo]

You don't have to use Notepad but you have to use a plain text editor. To associate .AVS files with Notepad right click on an AVS file, select Open With... If Notepad isn't in the list use the Browse button to navigate to it in C:\Windows\Notepad.exe. Be sure the "Always use the select program..." option is enabled.

[Alex_ander]

Isn't it necessary to keep the .avs associated with Avisynth?

[jagabo]

Isn't it necessary to keep the .avs associated with Avisynth?

No.

[Canon GL-2 Guy]

OK, I'm back from my sojourns to both the north and south for another day and it's time to pick up where we left off.

Bottom line: it's time to try the avisynth scripts provided earlier on.

COPIED AND PASTED:

WhateverSource() # 30i source
QTGMC() # convert to 60p
BilinearResize(720,480) # resize for DVD
SeparateFields() #convert to fields
SelectEvery(4,0,3) # select top and bottom fields
Weave() # weave them into frames


1) What if anything goes between the brackets?

2) Don't I need to put in the file name of the Lagarith .avi file somewhere into the equation/script?

3) Is this the variation I should try first?

Please, let's continue this discussion.

Thanks again

[jagabo]

1) Between the parenthesis you should enter any necessary arguments. Some commands don't require them, some do.

2) I used "WhatverSource()" to indicate you should use whatever source command necessary in AviSynth. For an AVI file you should use AviSource(). If the file is in the same folder as the AVS script you can use just the filename within the parenthesis: AviSource("filename.avi"). If the file is else where you need to use a full pathname: AviSource("C:\videos\filename.avi").

I suggest you start with something easy like the first script in post #11:

AviSource("filename.avi")
SeparateFields()
BilinearResize(720,240)
Weave()

Try the QTGMC() script once you get the basics down. Keep in mind that to use QTGMC() you have to download the QTGMC.AVSI script and put it in your AviSynth plugins folder. And it requires several other third party filters that you must download and install:

http://avisynth.org/mediawiki/QTGMC

[Canon GL-2 Guy]

Thanks,

Please allow me wrestle with this for the next few days and see how far I get (or not)