Best way to do this tweak and upconvert in the fewest steps

[brassplyer]

I want to take DV video - some of it 4:3, some of it 16:9, deinterlace with QTGMC cropping out the switching noise and blank areas on the side, convert to HuffYuv, clean with NeatVideo, use the Gunnar Thalin deshaker within VirtualDub, upconvert to 720P - pillarboxed in the case of the upconverted 4:3 video.

The process I currently envision would take 3 steps:

1. Load this avs script into Virtualdub

SetMTMode(3, 4)
FFMpegSource2("msourcefile.avi", atrack=-1)
AssumeBFF()
ConverttoYv12(interlaced=true)
SetMtMode(2) # uses less memory than mode 3
QTGMC (Preset="medium") #
Crop(8,0,-8,-8) <=== in the case of the 4:3 which is analog video and requires cropping.

and save to mysourcefile.avi as a 59.94 fps Huff file.

2. Process the resulting "mysourcefile.avi" with the Thalin deshaker within Virtualdub

3. Clean with Neatvideo and upconvert the resulting deshaken file to 1280x720 with a resize filter within Virtualdub.

The Thalin deshaker requires 2 passes so it seems it would prevent me from doing anything with the file in the same step as deshaking it. No interest in using the Avisynth stabilizer filter - from what I've seen it's a much bigger pain to use than the Thalin deshaker and doesn't work as well.

Thanks for all input


*Update - when I tried to do step 3 above - Neatvideo and resize in one step I eventually got a system out of memory error. It ended up saving about a minute of the video.

[sanlyn]

Crop(8,0,-8,-8) <=== in the case of the 4:3 which is analog video and requires cropping.

...which alters the aspect ratio of the original frame if black borders aren't added to restore the original frame size, which in turn will be exaggerated somewhat when the video is resampled.

However, because it appears that average viewers don't visually discriminate that well, and the overall style in processing these days is who-gives-a-crap, then a difference of 16 pixels in width (which becomes about 28 pixels in width after upsampling), could probably be dismissed. Most people watch videos stretched or squished even worse all the time and don't even know it.

[brassplyer]

Crop(8,0,-8,-8) <=== in the case of the 4:3 which is analog video and requires cropping.

...which alters the aspect ratio of the original frame if black borders aren't added to restore the original frame size, which in turn will be exaggerated somewhat when the video is resampled.

I was resizing to 720 x 480 with Virtualdub to retain the AR but now am doing it with the Resize command in Avisynth.

*Update - when I tried to do step 3 above - Neatvideo and resize in one step I eventually got a system out of memory error. It ended up saving about a minute of the video.

No surprises here. You might prevent that by running NV and resizing in two separate steps.

I find that I'm running into this problem even running NV alone.


According to this thread on the NV forums, apparently NV only accepts RGB video. The Avisynth script I'm using converts to Yv12, wondering if this could be causing a problem. However NV does process the video to a point. The reason I'm including the ConvertToYv12 command is that there are glitches in the video if I don't.

The avisynth script I'm using to deinterlace:

SetMemoryMax(768)
SetMTMode(3, 4)
FFMpegSource2("X:\Disney\Magic Kingdom\main st phil flag.avi", atrack=-1)
AssumeBFF()
ConverttoYv12(interlaced=true)
SetMtMode(2) # uses less memory than mode 3
QTGMC (Preset="medium") #
Crop(8,0,-8,-8)
LanczosResize(720, 480)

[poisondeathray]

According to this thread on the NV forums, apparently NV only accepts RGB video. The Avisynth script I'm using converts to Yv12, wondering if this could be causing a problem. However NV does process the video to a point. The reason I'm including the ConvertToYv12 command is that there are glitches in the video if I don't.

It doesn't matter, for 3 reasons. 1) vdub will convert to rgb anyways with most filters 2) deshaker export will be RGB anyways (you should set huffyuv to RGB for the export at that stage ) 3) you 've separated the steps already. neat video isn't "seeing" your original script - You've already applied to script to a lossless intermediate a few steps earlier

So split out the NV step to another lossless intermediate export, then upscale that in another step if you have memory problems, or use setmemorymax()

[sanlyn]

Crop(8,0,-8,-8) <=== in the case of the 4:3 which is analog video and requires cropping.

...which alters the aspect ratio of the original frame if black borders aren't added to restore the original frame size, which in turn will be exaggerated somewhat when the video is resampled.

I was resizing to 720 x 480 with Virtualdub to retain the AR but now am doing it with the Resize command in Avisynth.

Whether you do that in VirtualDub or in Avisynth, it's the same thing. Besides, Lanczos leaves edge artifacts IMO.

Crop(8,0,-8,-8)
LanczosResize(720, 480)

On display, your image will be stretched horizontally. Slightly. by about 12 pixels wide and maybe 1 or 2 high, but it won't be 4:3.

Crop(8,0,-8,-8).AddBorders(8,4,8,4)
Spline36Resize(720, 480)

If you upsize a second time to 1440x1080, you'll start seeing quality loss.

BTW, how do you get bottom-border head switching noise out of DV? If you captured VHS to lossy DV, then encode again, you've lost something more. Ever heard of encoding once and resizing once? I don't know...I guess it's more efficient to do it all twice.