How does VirtualDub sucessfully downscale HD to SD widescreen?

[Canon GL-2 Guy]

VirtualDub is giving me a trying workout

I'm endeavoring to leave no stone unturned with no path unpursued in my quest to successfully transform HDV 1080 60i 16:9 master footage to SD NTSC widescreen video (853 by 480).

According to a thread that I accidently found in this forum, one of the possible/doable solutions in regards to the downscaling part of the equation with at least some degree of quality is Virtual Dub.

1) Do I want the MPEG version or the AMD-I7 (et. al) version or does it really make any difference at all?

I exported a 7-8 min test clip out of CS 5.5 with AME in 3 different flavors to a Lagarith AVI Rightly or wrongly "the decode while upsampling" box was not checked. Was that proper? Assume 24 bit. Lanzcos 3 filter applied.

The only difference was the colorspace: (RGB and YV12 and YUY2)

The only one Premiere Pro plays is RGB, the others are garbage and/or/ displays a black screen on playback.

When these files were input into VB after resizing what I got was:

1) Output file size remained the same as the original (1440 by 1080). Color/audio ok.

2) Black screen with audio only.

3) Very limited color (grey, black and white striped video on a slant!) ZEBRA.

Which presently leaves me at a loss struggling to get a handle on how to proceed.

Please, would any of you be so kind to relay your personal settings and/or attach screenshots of the settings so I can get back on track and on target?

As usual, any advice or pointer are highly welcomed.

Thanks in advance and have a enjoyable holiday weekend.

[poisondeathray]

Upgrade to the latest lagarith, there is a bug fix for Adobe issued about a week ago

Version 1.3.26 released on 09-25-2011
Fixed a buffer overrun in the decoder that caused crashes with Adobe products.

NTSC is 720x480, not 853x480 . 853x480 is the square pixel equivalent, but you want 720x480

You didn't specify if you are going back into premiere or what your goal is or what stage you're doing. Are these already edited? Are you going to DVD with another encoder ? Do you plan to bring these back into premiere ?

I've already stated this in your other thread, but editing it as HDV in an HDV timeline has many "pros" and very few "cons". I'm not sure what you're doing... You're wasting time, harddrive space, and possibly losing quality unecessarily from colorspace conversions by taking so many steps

When these files were input into VB after resizing what I got was:

1) Output file size remained the same as the original (1440 by 1080).

I don't understand what you're saying here. If they have already been downscaled, how can it be the same as the original ?

[Canon GL-2 Guy]

Hi PoisonDeathRay,

Please, let's back track a little and consider the/my particular workflow picture.

1) The full editing process has already been completed, rendered and finalized.

2) That HDV project is exported direct from the project timeline and saved as a Lagarith AVI file.
2b) I need to know what type of Lagarith file is best suited for Prem Pro for eventual encoding to MPEG-2 dvd spec.

3) As we all know, Premiere Pro strikes out heavily for downscaling quality which is the underlying foundation and premise for this post.

4) The next step is to downscale that Lagarith file using VirtualDub where I'm having technical difficulties.

5) Obviously, I'm setting up the parameters wrong in VB which is yielding terrible results. Need specific guidance/screenshots here.

6) If I still end up with 1440 by 1080 after attempting to "downsize", what I tried did not work !

Many thanks for the bugfix link which I'll put into action very quickly.

Please, let's continue with this discussion.

Thank you sir.

[poisondeathray]

First, the vdub interlaced downscale isn't the greatest because it simply separates fields then resizes them using whatever algorithm you choose e.g. Lanczos. What you should be doing is smart bobbing the footage, resizing, then re-interlacing as jagabo described in the other thread. The vdub workflow is certainly better than Adobe's interlaced scaling , but still... (also CS5 with MRQ and CUDA might produce "acceptable" results for some people now)

If you've exported a 1440x1080 60i lagarith AVI from AME, it's important that it's NOT the YV12 variant (because vdub doesn't handle interlaced YV12 correctly, it upsamples as progressive and you will get chroma artifacts) . If you've used filters like color correction in your edits, I would export lagarith in RGB mode (checkmark always suggest RGB box)

In the resize dialog, check disable aspect ratio first, then enter 720x480, select lanczos3 from the drop down box, then checkmark interlaced. Push preview to see if it looks ok, then apply and ok

Exporting lagarith in RGB mode after applying the vdub resize should work in premiere, but be aware there are other MPEG2 encoders that some people may prefer over the mainconcept based MPEG2 encoder bundled with Premiere eg. HCEnc


If you're putting this back into premiere, you usually have to interpret the aspect ratio (most intermediates are "seen" by premiere as square pixel 1:1 PAR, but NTSC 720x480 doesn't use square pixels, so right click the re-imported clip and interpret the proper 16:9 aspect ratio)

If you use avisynth you can skip a few steps and don't need some of these intermediate files or stages . That's what it does - frameserve. You can also do a better resize using scripts

[racer-x]

I rather like the Virtualdub resizing algorithm for 1080i > 480i and have made it my standard tool for this job. I allays shoot and edit in full HD and output to Blu-Ray, however some people still use need SD dvd versions. I use Vegas to edit and I frameserve the timeline into Virtualdub, apply the resize filter and then frameserve into TMPGEnc. The quality is very exceptable and no aliasing issues with the HD > SD resize that is prevalent in many other methods.

I do as PoisonDeathRay suggested, only I resize to 704 x 480 and letter box out to 720 x 480:

[poisondeathray]

Letterboxed 704 is probably more correct for 16:9 DVD , but there are examples of retail authoring houses that do it both ways

Adobe follows the convention of using a PAR of 40:33 (1.2121) for 16:9 NTSC DVD , so it bases it's calculations on 704px width of the active image. If you use AME, you will notice 16:9 DVD's you make will have letterboxing

Full frame would use 32:27 , based on 720px width of the active image

The other thing you should do is convert from Rec709 => Rec601 for SD

[jagabo]

I covered this in the OP's other thread but I'll repeat it here. VirtualDub's method is one of the crudest. That VirtualDub resize is the equivalent of

SeparateFields()
LanczosResize(704,240)
Weave()
AddBorders(8,0,8,0)

in AviSynth. That will lead to ringing and blurring on sharp edges. Fortunately, most camcorder video isn't so sharp so it's not too visible.

1080p source:


480p result (upscaled 3x, nearest neighbor):


Notice the white halos and the black line beneath the border between the white and gray. Actually, most of that is due to oversharpening by Lanczos3. You can alleviate most of it by using Precise Bilinear instead. Of course, that will result in a less sharp picture.

A better method is to use a good smart bobber, resize, then reinterlace. Unfortunately, that can be very slow.

It's not really necessary to resize to 704 then pad to 720. One can resize to 720 directly. The DVD spec refers to the MPEG 2 spec regarding aspect ratio. The MPEG 2 spec clearly states that the full frame comprises the flagged aspect ratio. Of course that's at odds with with the BT/ITU spec for digitizing analog video. You could avoid the conflicting specs and resize to 704x480 and leave it there. 704x480 is compliant with the DVD spec.

[poisondeathray]

One reason why someone might prefer 720 encoded width (either letterboxed or not) is that it is compatibile with SD blu-ray, but 704 is not . But then again, why would someone make SD blu-ray when they have an HD master to go from...

[racer-x]

Crude or not, it looks darn good with my Canon HG20 (AVCHD) source clips.............

I do agree that using a Smart Bobber, Resizing, then re-interlacing will give a sharper overall picture. However I do HD > SD resize on sports footage and this method tends to show a lot of interline twitter during panning where there are thin horizontal lines in the background.

Virtualdub resize won't do that, but it is a little less sharp. It's also very fast and the quality is very acceptable.