Any Faster Way for De-interlacing Using QTGMC?

[poisondeathray]

crop and resize in the script ? or in vdub? Probably faster in the script

quick test:

QTGMC + crop + lanczos resize all in avs script was about 2.3-2.5x faster than QTGMC only avs but doing the same crop and lanczos resize operations in vdub2. Vapoursynth version ~2.7-2.8x faster (when you add more than 1 filter in the filter chain, vapoursynth tends to be faster, the treading model is more efficient). All measured using vdub2 analysis pass .

(And if you're upscaling, you should also adjust for SD 601 => HD 709 colors, by using colormatrix or similar)

[dellsam34]

I'm more comfortable with vdub because I can see the frame being worked on, Avisynth won't show me anything. Im assuming use 709 when I encode to h.264.

[poisondeathray]

I'm more comfortable with vdub because I can see the frame being worked on, Avisynth won't show me anything. Im assuming use 709 when I encode to h.264.

Do you prefer it more than processing at 1/2 speed ? Which Christmas? This or next year's ?

You can use vdub2's script editor, push f2 to refresh

Avspmod has a preview and interactive crop editor( where you can drag the left, top, right bottom sides with a mouse) . It also has sliders, which are user programmable, so you can adjust filters with a mouse. It's the closest thing to a GUI for avisynth

If you don't adjust the colors, they will be the same as original (SD, 601) . Just encoding using h264 does not change actual colors. YUV values will be the same. Just flagging the 709 values does not change the actual colors (it's just metadata). But playback devices, TV's, software players assume 709 for "HD" resolution (they are converted to RGB for display using 709) , so the colors will be off if you upscale and don't make the adjustment.

colormatrix(mode="rec.601->rec.709", clamp=0)

[johnmeyer]

If you capture SD using "uncompressed" (not recommend, and no reason to ever do so), then your disk subsystem could become a bottleneck. Capture instead using a lossless codec like Lagarith or HuffYUV.

[jagabo]

You can use vdub2's script editor, push f2 to refresh

F2 when the main VirtualDub window has focus, F5 when the editor has focus.

[dellsam34]

Do you prefer it more than processing at 1/2 speed ? Which Christmas? This or next year's ?

Lol.. That's what I'm trying to figure out, if I I need a bigger processor than I will have to build a custom mini form factor desktop for just this purpose, The last time I've built a computer was 15 years ago so I need a lot of catching up with newer technology.

[poisondeathray]

You can use vdub2's script editor, push f2 to refresh

F2 when the main VirtualDub window has focus, F5 when the editor has focus.

oops, thanks, I don't really use it

avspmod is f5 too. Tabs are the best feature that convince me to use avspmod instead of vdub2

Do you prefer it more than processing at 1/2 speed ? Which Christmas? This or next year's ?

Lol.. That's what I'm trying to figure out, if I I need a bigger processor than I will have to build a custom mini form factor desktop for just this purpose, The last time I've built a computer was 15 years ago so I need a lot of catching up with newer technology.

It's not that hard to move the crop and resize into the script. A shiny new processor will still process much slower than it could, if you use vdub to do the crop and resize steps. More than 2x faster is quite significant

[manono]

Here is the above sample on youtube after being processed:
https://youtu.be/HEsS3xmaUeA

That's the 'finished product'? You can't tell just by looking how badly blown the whites are?

If you find QTGMC too slow, then maybe switch to:

yadifmod(mode=1, edeint=nnedi3(field=-2))

Nearly as good and a helluva lot faster. And you really should be doing all the filtering in AviSynth.

[jagabo]

I'm more comfortable with vdub because I can see the frame being worked on, Avisynth won't show me anything. Im assuming use 709 when I encode to h.264.

You can easily transfer the values from VirtualDub to the script.

[dellsam34]

That's the 'finished product'? You can't tell just by looking how badly blown the whites are?.

No, not even remotely close, still working on a finished product here, don't you see? That youtube video shows the cropping. There is nothing I can do about clipped levels, it's source related and it's inconsistent, If I have to go that route I will finish by Christmas 2030.

Poisondeathray, I will try Avspmod when I get home.

[dellsam34]

I opened AVSpmod from the extracted folder and loaded a script and it asks me for an external program, What external program should I use? I hope it is not going to be another episode of QTGMC and it's dependencies, it was a nightmare.

Also when choosing to resize after loading the cropping script it won't let me select 1440x1080 it wants 1440x1088, no matter what I do.

[Nands67]

Try this, SetFilterMTMode("QTGMC", 4) which should be ok for 8 cores

[dellsam34]

Oh I see, Vdub2 as an external program, Okay now I loaded QTGMC in fast mode and prefetch3, crop filter and resize filter all 3 averaging 20 fps, not bad. Now guys I need a good encoding script for the resulting 1440x1080 NTSC video into h.264 I would like to keep 4:2:2 and audio @ 192Kbps, That way I can load all scripts in one go.

Poisondeathray thank you so much for the AVSpmod suggestion, it was the missing tool for my scripting needs, it's the best Christmas gift.

[Nands67]

Sorry, and this,
QTGMC(preset="fast", TR2=3, EdiThreads=2)
Prefetch(6)

[dellsam34]

I don't think I have more than 4 threads, It's a quad core computer.

[Nands67]

Create a batch file in the same folder as your AVS file with this script inside.
ffmpeg64 -i "test.avs" -c:v libx264 -preset slow -crf 22 -filter:v fps=fps=30 -c:a aac F:\Video_Output\test.mp4
you need to change the file names and drive letters to suit.
Then run the batch file from CMD prompt

[Nands67]

Cntrl alt del, task manager, performance, will show how many cores and logical processes.
Set the mode to half the number of logical processes, ie 8 logical processors =4, SetFilterMTMode("QTGMC", 4)
Set edithreads to half the number of cores, ie 4 cores edithreads=2

[dellsam34]

We've gone thru this already in the first few posts, Check post #11.

[lordsmurf]

Well, They want the highest quality possible

You must understand that QTGMC "slower" isn't higher quality than "faster". It's just different.
And slower tends to blur. That's why I never use it.

I/O may be in play here.
- Are you going from Drive A to Drive B?
- Note that SSD can be better than HDDs, though not always.

Same-drive read/write will massively slow down processing.

All deference to selur, but I don't like the default Hybrid x264 encode settings. I've discussed my settings before on digitalFAQ.com/forum a few times. Did you see those posts? For example, 422, longer GOP, pull back to CRF 15 for HQ from VHS sources. Advanced mode, not the newbie/crippled mode.

[Nands67]

You may have to enable hyperthreading
You can increase the numbers untill 100% cpu usage.
If you go too high it will slow down.

[Nands67]

Yes i agree, have the write file to separate HDD and the file read/processing on the SSD

[Nands67]

You could also set the cpu priority for avspmod, ffmpeg, virtualdub etc
I think there is a way to set it permanent but not sure.
Setting in task manager is not permanent so you have to do it each time.

[dellsam34]

All deference to selur, but I don't like the default Hybrid x264 encode settings. I've discussed my settings before on digitalFAQ.com/forum a few times. Did you see those posts? For example, 422, longer GOP, pull back to CRF 15 for HQ from VHS sources. Advanced mode, not the newbie/crippled mode.

No, I did not because I've never used Hybrid before, I used ffmpeg on de-interlaced and interlaced SD materials depends on the situation, This project requires me to encode from 1440x1080 de-interlaced files, I will check them out and start using hybrid if it is more convenient.

And no it's not the hard drive we ruled that out already, An SSD cannot choke on 25 fps at 5% usage, it's the CPU.

[manono]

No, not even remotely close, still working on a finished product here, don't you see?

Oh, okay. When you wrote, "after being processed", I thought you meant after being processed by you, not realizing you meant processed by YouTube. Even if the contrast values are inconsistent, you can still lower them for the whole thing quite easily. As soon as I began to watch that YouTube video, my eyes hurt. Then I downloaded it to examine it and the contrast values are awful. You couldn't adjust them a bit before capturing? Because a lot of the whites can't be recovered, they're so bad.

If you really want the 'highest quality possible', that ain't it.

[Selur]

some general comments:
around QTGMC:
a. Vapoursynth is usually 2+ times faster than Avisynth when using QTGMC
b. Avisynth Neo + KTGMC is 7+ times faster than normal Avisynth using QTGMC
c. Vapoursynth for me is always faster when using multiple filters; it really can help to use nnedi3cl instead of nnedi3 to free some cpu time
-> If you only use QTGMC and you have an nvidia card looking into Avisynth Neo and KTGMC really might be worth it.
about processing speed: during encoding hdds are usually not the slowing down anything (assuming you don't process uncompressed content, which seems just silly)
If you run any anti virus or firewall stuff on your system, make sure that they don't touch your input/output/temp folders during processing, since that really can slow down things.
Especially when you build process chains of multiple tools like using Avisynth/Vapoursynth for filtering, FFmpeg/VirtualDub for processing the video and something else for encoding.

---

All deference to selur, but I don't like the default Hybrid x264 encode settings.

Yup, those are meant as defaults that should work fine for most content, so changing them is recommended if you have the time and motivation to read up on H.264 and x264.

Cu Selur

[dellsam34]

Oh, okay. When you wrote, "after being processed", I thought you meant after being processed by you, not realizing you meant processed by YouTube. Even if the contrast values are inconsistent, you can still lower them for the whole thing quite easily. As soon as I began to watch that YouTube video, my eyes hurt. Then I downloaded it to examine it and the contrast values are awful. You couldn't adjust them a bit before capturing? Because a lot of the whites can't be recovered, they're so bad.

If you really want the 'highest quality possible', that ain't it.

Unfortunately I can't, there are dark scenes that will suffer, Overall the BE75 will alert me if there is level clipping during capturing that is not source related, If it's the shooter's fault it ain't my problem, I was paid to capture not to restore.

[abolibibelot]

Just for laughs... can't resist... a hilarious rant about camcorders from dearly missed George Carlin...
“And camcorders. Here is technology gone berserk. Everywhere you go now there’s some dick, some yo-yo, some putz with a camcorder, and he's going to tape... everything! Doesn’t anyone in this country just stop and look at things anymore, sort of take them in, maybe even… remember them? Is that such a strange notion? Does experience have to be documented and brought home and saved on a shelf? And do people really watch this shit? Are people’s lives so bankrupt they sit at home looking at things they already did? These guys are so intense, you know? It’s always guys; they won’t let women touch the cameras, it’s a highly technical skill — look through a hole, push on a button, big skill. And they all think they’re Federico Fellini. You ever see ’em here with this shit? Low angles, zooms and pans, and it’s the same ugly three children in every goddamn shot. All the George Lucas magic in Hollywood is not gonna change the unfortunate genetic configuration on the faces of these children. Keep these unfortunate youngsters out of public view!”

[poisondeathray]

a. Vapoursynth is usually 2+ times faster than Avisynth when using QTGMC

In CPU only mode, QTGMC only, I find avs+ mt version a hair faster . opencl=True will depend on your GPU, it can be even slower than False. But NNEDI3 is one of the large bottlenecks for QTGMC . If you can speed up that step, QTGMC will speed up

It also depends on the CPU. The vpy version uses ZNEDI3, which is significantly faster if you have newer model with AVX512

b. Avisynth Neo + KTGMC is 7+ times faster than normal Avisynth using QTGMC

Depends on the GPU. For some people it's slower than CPU avisynth+ MT

c. Vapoursynth for me is always faster when using multiple filters; it really can help to use nnedi3cl instead of nnedi3 to free some cpu time

Definitely . The threading model is much better

Another issue with avs+ QTGMC is sometimes the ideal prefetch values is actually some other number, depending on the script . VPY always gets it automatically optimal without having to do some tests or avsmeter64


Also there are slight quality differences between all of them. Not necessarily better or worse. eg. The prescreener for NNEDI3CL is different than NNEDI3CL and KNNEDI3

[dellsam34]

Just for laughs... can't resist... a hilarious rant about camcorders from dearly missed George Carlin...
“And camcorders. Here is technology gone berserk. Everywhere you go now there’s some dick, some yo-yo, some putz with a camcorder, and he's going to tape... everything! Doesn’t anyone in this country just stop and look at things anymore, sort of take them in, maybe even… remember them? Is that such a strange notion? Does experience have to be documented and brought home and saved on a shelf? And do people really watch this shit? Are people’s lives so bankrupt they sit at home looking at things they already did? These guys are so intense, you know? It’s always guys; they won’t let women touch the cameras, it’s a highly technical skill — look through a hole, push on a button, big skill. And they all think they’re Federico Fellini. You ever see ’em here with this shit? Low angles, zooms and pans, and it’s the same ugly three children in every goddamn shot. All the George Lucas magic in Hollywood is not gonna change the unfortunate genetic configuration on the faces of these children. Keep these unfortunate youngsters out of public view!”

As off topic as it is I L.M.A.O. It has gotten way worse now with cell phones and everyone doing it not just men.

[dellsam34]

Poisondeathray what is a good script for encoding those HuffYUV de-interlaced 59.94 fps 1440x1080 files @ 4:2:2 and 192Kbps audio?