newbie to AVISynth & VirtualDub-need help with IVTC & deinterlace via QTGMC

[jagabo]

That was originally telecined film but it has been blend deinterlaced (the two fields were blurred together). You'll need the FixBlendIVTC() filter for that.

http://avisynth.nl/index.php/FixBlendIVTC

Get the "2017" version. Then the basic fix is:

import("C:\Program Files (x86)\AviSynth\plugins\FixBlendIVTC.avs") # adjust path as necessary
AviSource("YUY2 test.avi") 
ConvertToYV12(interlaced=true)
FixBlendIVTC()
TDecimate()

[clashradio]

I captured the same clip only this time using the BlackMagic Intensity Shuttle and I can see the interlacing in the raw video, which I don't see in the VirtualDub capture. Do you think somehow when I'm capturing in VirtualDub I have some mode set to Progressive? I do see a Preview field mode, which is set to Progressive. I wasn't sure if that should be set to Frames, or Weave even/odd?

[jagabo]

I don't think the preview mode makes a difference in VirtualDub's captures.

The new cap with the Shuttle looks fine. Just use TFM() and TDecimate().

[clashradio]

Ya I was thinking the Preview mode in VirtualDub didn't matter.

If the Shuttle cap seems to be interlaced, why do my VirtualDub caps appear to be Progressive, or blend deinterlaced?

[jagabo]

I'm not absolutely sure about VirtualDub's preview setting. You should try different settings to see if there's any difference. Also make sure Video -> Filter Chain is disabled -- or at least you're deinterlacing there. And make sure Video -> Vertical Reduction is set to None.

[clashradio]

I figured it out after checking your suggestions, which were disabled already.

The problem was the capture program for the Osprey 827e was set to deinterlace upon capture. I didn't catch it because the first LD capture tests were video-sourced. And I'm not sure how that would matter since I'm using VirtualDub for capturing.

[LMotlow]

VirtualDub doesn't have capture drivers. It simply hooks into your capture device drivers and they do what they're programmed to do.

[clashradio]

jagabo: I'm having trouble determining what script I should use on this particular file. I tried IVTC & QTGMC. The later is better, but looks like there are a few missing frames. Below is the script I'm using.

AviSource("input.avi")
AssumeTFF()
ConvertToYV12(interlaced=true)
QTGMC()
SRestore(frate=23.976)

[manono]

Nearly 400 MB? How about 10 well-chosen seconds? And is this the source or your reencode? Only the source will be remotely useful.

.I tried IVTC & QTGMC.

You don't just blindly try stuff. You study your source and then put on the correct filters.

[clashradio]

Sorry about the 400mb. I wanted to find a section with lots of motion and w/o re-watching the entire concert again this is the best I could find. This is a direct new capture using YUY2, 4:2:2. I'm still having trouble studying the source to determine the best method.

[manono]

Sorry about the 400mb. I wanted to find a section with lots of motion and w/o re-watching the entire concert again this is the best I could find.

I don't see a smaller piece. However, if the choice is between whether to IVTC or to bob, that's easy. If it was shot using video cameras, every frame is interlaced during movement and if it has to be made progressive, QTGMC would be the best way to do it.

If it was shot on film then in every 5-frame cycle, 3 frames are progressive and 2 are interlaced. That requires an IVTC. This assumes it's not converted from one standard to anopther (PAL<->NTSC and field blended). If every frame is progressive, then nothing has to be done.

You create a script with no filtering, open it in VDub, and have a look during a section with movement. You said this is from a concert so finding motion scenes shouldn't be hard.

[jagabo]

Looks like PAL to NTSC with field blending:

AviSource("jagabo.avi") 
AssumeTFF()
ConvertToYV12(interlaced=true)
QTGMC(preset="fast")
SRestore(25)

[LMotlow]

Nearly 400 MB?

And it's uncompressed YUY2, which made it even bigger. The O.P. doesn't know about lossless compressors? I compressed the sample with YUY2 Lagarith and got 119MB.

[clashradio]

jagabo: I'm having an issue with what appears to be improper IVTC? Please take a look at frames 33 & 41. You can see some interlacing in Betty's face. These frames correspond with after the IVTC has been performed. Code I used is:

AviSource("YUY2 422.avi")
AssumeTFF()
TFM()
TDecimate()

I also discovered some different quality settings since my last post, which did seem to help. I know every capture program is different, but for Video Scaling Algorithm I used: Lanczos. Options are (from lowest to highest) Nearest Neighbor, Linear, Cubic, Lanczos. My original caps were with Linear.

I also lowered my Video Buffers to 10 (default is 50). My original caps were set to 50.

Should I be using a lossless compressor in general or just for clips meant to be uploaded? I've got the storage space so I've been using Uncompressed RGB/YCbCr vs. FFMPEG/Huffyuv/Lagarith etc.

[jagabo]

Your script look to be right. But please send clips with more motion. It's much harder to check clips with so little motion. There are some artifacts in that cap that concern me a bit though. After a Bob(), at fields 15 and 16 there is some ghosting from the previous field. Also at fields 158 and 159. Both of these are at shot changes and probably have something to do with the comb filter. After TFM().TDecimate() you can see the ghosting at frames 6 and 63.

You should be using lossless compression all the time. Generally, modern systems run out of disk bandwidth before they run out of CPU when capturing SD video. Using the CPU to compress the video to 1/3 the size gives you much more headroom to keep from dropping frames.

I'm not sure whether or not the scaling algorithms are being used during your caps. I recommend you compare Linear (least sharp) and Lanczos (most sharp) and see if there's any difference. They may only apply if your are downscaling the frame while capturing (like scaling to 640x480 to get the right aspect ratio). Generally, given the low horizontal resolution of VHS you don't want to use very sharp scaling algorithms (Nearest Neighbor, Cubic, Lanczos) because they will accentuate noise without sharpening the underlying image. Linear will actually give you a little noise reduction.

[clashradio]

Would you prefer I do not have the 3D comb filter in the chain for these cap tests? (Yes the ghosting in those frames I believe is from the comb filter, known as checkerboarding).

Ever since you said not to cap with the audio preview on, I haven't had one dropped frame. I figured uncompressed is better than lossless? If you recommend lossless, which option? I'm hoping not Huffyuv as I can't seem to select YUY2 like you can with Lagarith. My scratch drive is a Samsung 970 NVMe M.2 SSD, so it's got the speed.

Ya I was a little confused about the Video Scaling Algorithm dropdown. Those options are listed as resizing methods over at https://old.medialooks.com/video_scaler/. Osprey support says they use the DirectShow API for their drivers. I'm capturing at "same as input signal" as far as the option "default video input size", so hopefully no resizing is going on.

So you're saying you recommended Linear vs. Lanczos due to the low rez of laserdiscs?

[clashradio]

Here is a new clip with lots of motion, captured using Lagarith, 4:2:2 YCbCr (YUYV, YUY2) without the external 3D comb filter.

[jagabo]

Wow, that's much better. Less noise at vertical edges, more fine detail, no dot crawl artifacts. Needs some levels adjustments. Basic IVTC script:

AviSource("no comb filter.avi") 
AssumeTFF()
TFM()
TDecimate()

[clashradio]

What's weird is you say it has less noise and more detail. I can't tell if this is because of the lack of the comb filter, or because I captured using Lagarith? Because for composite video I thought you needed a comb filter of some sort due to the y/c separation? Or do you recommend a 2D? In all the research I've been doing, everyone says 3D due to 3D has more vertical resolution than 2D, plus 3D handles the orb pattern better.

Attached is a grabbed frame from the Video Essentials laserdisc Snell & Wilcox orb. Image on Left=no comb filter, Right=3D comb filter.

[jagabo]

What's weird is you say it has less noise and more detail. I can't tell if this is because of the lack of the comb filter, or because I captured using Lagarith?

Lagarith is lossless. It will not cause an increase in detail or noise.

All along (here and via PM) I've been saying you should try all the different scenarios and examine the results yourself. Capture the same section of video with each scenario so you can compare directly. See what effect each piece of equipment has and weigh the advantages and disadvantages. Also be aware that something that works well with one source may not work well with another. So you have to be open to using different setups for different videos. Though with laserdisc the variation is much smaller than with VHS.