Jerky footage after restoration attempt...

[low-fat-al]

I was trying to restore an old music video ripped from DVD.
I attached a 10 second sample for help in this.

After using the following script:

-----------------------------------------------
SetMTMode(5, 4)

mpeg2source("D:\California Love.d2v",cpu=6)\

SetMTMode(2)

ConvertToYV12()

Checkmate(24)
DeGrainMedian(limitY=5,limitUV=7,mode=0)

QTGMC(preset="slower",sharpness=0.3)

srestore(frate=23.976)

---------------------------------------------------------

I put in VirtualDub and cropped and resized for 640x360. I also used NeatVideo to clean up some grainy parts.
It looks great!
The problem is that there are parts that seem to look like it is not smooth motion.
Watch the 10 second sample of the final export and 4 seconds in on that scene change.... it appears jerky.... then scene change and it goes back to normal.



Here is a link to the same 10 second segment CineForm AVI final export out of VirtualDub.
https://dl.dropboxusercontent.com/u/30532277/California%20Love.avi

[sanlyn]

Offhand:

-Frames 0-10 have no B frames. Only I and P.

-Telecined video was deintelaced.

I put in VirtualDub and cropped and resized for 640x360. I also used NeatVideo to clean up some grainy parts.
It looks great!

-Reds oversaturated.

-No clean whites (that's not a pun )

-Darks crushed.

[low-fat-al]

Alright....

Suggestions?

[sanlyn]

DVD comes into the world as either interlaced or progressive/with 3:2 pulldown. If so, then QTGMC as used would double the frame rate (59.94 FPS progressive). Yet your srestore statement wants 23.972 fps output. So, a few things don't match up.

The avi appears to be progressive, 23.976 FPS.

[manono]

Alright....

Suggestions?

IVTC it. It's not blended so you don't need SRestore. Where the cadence is broken at scene changes there are sometimes, but rarely, some minor problems. And as sanlyn says, there's lots else wrong with this thing. Your 'result' is soft as mush.

It plays smoothly for me. Maybe that Cineform AVI is taxing your computer? There is a part in the middle, though, where every other frame is duped. Don't know how you managed that.

[sanlyn]

Since you already have an unprocessed original vob or mpg, you can use DGIndex to cut an m2v of the same segment so people can have look at the unprocessed source.

[poisondeathray]

-no clean whites (that's not a pun )

lol

[manono]

Since you already have an unprocessed original vob or mpg, you can use DGIndex to cut an m2v of the same segment so people can have look at the unprocessed source.

sanlyn, he's got an MPG in the first post. I assume that's unprocessed and from the source.

[sanlyn]

I didn't get that from the first post, but I'll re-read it.

read-read-read....

I'll assume the mpg was a cut from the original source. I'll also assumer it had some interlace/telecine problems, and will assume srestore was used to clean up whatever QTGMC delivered. Looks like the word "rip" threw me off. because on closer look it appears the DVD is home made. Why would someone say they "ripped" a DVD with no encryption? I'll assume something was used to copy and join VOB's. Oh, well....so much for the word "rip", which now is used for everything from transcoding/decrypting video to poaching eggs.

I'll assume that the mpg is a piece of the original. Back to the drawing board.

[low-fat-al]

Alrighty...

A few things...

I am still wrapping my head around the specifics of how some of the filters work with AVISynth.... so I had no idea that "cpu=6" is for strong deblocking.

The video source and 10 second segment is an untouched VOB rip from a DVD of various 2Pac videos. So rip in the sense of being ripped from the DVD....

I ,of course, used the full length for my situation... just needed to cut a segment for this post.
The AVI is fine... I even thought it probably was the Codec used... so I exported as a different codec and it did the same thing on the same parts.

The final video output is intended to be used for a video edit using Premiere Pro... so I need it in CineForm and won't export in any lower color depth than RGB24... it performs best and is in a small enough size as well.

I am removing some of the overkill parts and toning down the NeatVideo filter a bit.... but that wasn't my main issue... it was the jerky video.
I will export with new settings and hope it works fine.

[sanlyn]

Test and tweak your filtering changes with a few seconds of video, not the whole vido. The filter setting in NeatVido that has the most visible oversmoothing effect is the high-frequencyl setting in the advanced interface. It affects mostly fine-grain stuff -- which is where most of the so-called "detail" is in home-grown videos.

I ran the following on the mpg and don't see any jerky playback: Replace QGTMC and sRestore with this:

AssumeTFF().TFM(order=1).TDecimate()

I didn't have time tonight to address specific noise or dot crawl issues, but anti-dot and any anti-alias should usually be run before inverse-telecine or deinterlacing. You shouldn't need deinterlacing here, AFAIK, as the TFM/TDecimate routine gives you 23.97 FPS progressive. If you still want to resize: denoise and cleanup first, resize after. Avisynth has better resizing tools than VirtualDub.

I'm certain other readers are looking at the vids and will have more ideas. I hate dot crawl. The cleaners for it just seem to ruin everything.

[hech54]

Most likely it is a bootleg or homemade DVD. Just because it came from "a DVD" doesn't mean someone didn't try an amateur cleanup/conversion job from a crappy source. Eagle Rock has put out some very dubious crap over the years.

[low-fat-al]

If I can locate a better copy I would use it. Gonna experiment a bit and see what I can come up with.
Thanks for the replies.

[low-fat-al]

Odd...

The QTGMC along with srestore(frate=23.976) did something weird.

I redid the script with using:

TDeint()
Decimate(5)

and it advanced frame by frame like it supposed to in virtualdub... no repeat frames.

I put back in:

QTGMC(preset="slow",sharpness=0.3)

srestore(frate=23.976)

and in the same section I found where motion was jerky... every frame was duplicated.... so it was one frame... dup.... next frame... dup... next frame ... dup...

Am I using the QTGMC wrong to make it decimate every fifth frame like Decimate(5) ??

Any experts out there?

[manono]

TDeint()
Decimate(5)

and it advanced frame by frame like it supposed to in virtualdub... no repeat frames.

Yeah, but you deinterlaced it. I said to IVTC it. That might go:

TFM()
TDecimate()

Am I using the QTGMC wrong...

Yes, since you shouldn't be using it at all (or SRestore), except maybe as the post-processor. But the sample didn't need any post-processing

[sanlyn]

I went with TFM/Decimate. Worked OK (after I cleaned up my own dumb coding mistakes elsewhere in my script!). The mpg source has a lot of chroma noise, clumpy grain, and some aliasing. I see reddish grain and mottling in the black letterbox bars. The mpg seems to have been made from tape (head switching noise along the bottom).

[sanlyn]

Working with the mpg alittle this morning, I can see why the O.P. used some strong filtering. Dark areas have foggy, blocky noise that that generate banding if it's not overly smoothed out of shape. the dot crawl and jaggies are really stubborn, too. It looks as if the deepest dark areas are crushed in the original, so there's very little "down there" that doesn't have a lot of noise and data loss.

[sanlyn]

Color and levels a real problem/headache/hassle. They're a mess. They change with every camera shot (thank tape source for that!). You could tweak and play forever and make it look however you want. Denoising, not that much of a problem. Still has some rough edges.

Used this script. Finished off in VirtualDub with color and medium-power NeatVideo. Still not quite what I'd want, but everyone wants something different. But the original is way too red, crushed, highlights blown out. Couldn't retrieve all of it. There are a couple of dirty pixels along the side borders; one could crop them off.

MPEG2Source(vidpath+"California Love.d2v")
Crop(4,66,-8,-66)
SmoothLevels(14, 1.0, 255, 14, 245,dither=100,chroma=150)
AddBorders(4,66,8,66)

checkmate(24)
TFM()
TDecimate()
santiag()
MCTemporalDenoise(settings="high",sigma=5,AA=true,useEEDI2=false,enhance=true,\
  deblock=true,edgeclean=true)
GradFun3(thr=1.0,mask=0)
AddGrainC(1.5, 1.5)
Dither_convert_yuv_to_rgb(lsb_in=false,tv_range=true,cplace="MPEG2",output="rgb32",matrix="601")
Crop(4,66,-8,-66)
Spline36Resize(640,360)

If you wanted to go standard def BluRay, etc., this clip is not compliant.

[low-fat-al]

Hey...

I want to try your script but I can't seem to locate "santiag" ....

[manono]

Santiag:

https://forum.videohelp.com/threads/355015-having-trouble-finding-Santiag-without-the-usual-bullshit

[low-fat-al]

ok... now I am getting an this error:

Script error: there is no function named "grandfun2db"
(MCTemporalDenoise.v1.4.s0.avsi, line 700)

I have "gradfun2db.dll" and "GradFun2DBmod.v1.5.avsi" in plugins folder for AVISynth...

Am I doing something wrong?

[sanlyn]

The attached ZIP file contains the plugins and functions used by MCTemporalDenoise. The ZIP has 3 folders inside:

1. Plugins for the avisynth plugins folder.
2. Documentation for most of the stuff.
3. FFTW3.DLL (for System32 or for SysWOW64 if using 64-bit Windows). This is not an Avisynth plugin.

The conversion to RGB32 is done with the Dither plugin. You'll be up all night figuring that one out, so just use "ConvertToRGB32(matrix="Rec601", interlaced=false)". GradFun3() mis also in the dither package, so you can replace it with GradFun2DBmod(thr=1.8,mask=false). GradFun2DBMod is in the attached ZIP.

I don't care for most of the color stuff in the mkv I posted, but it's getting late around here. WIll try to clean it up tomorrow. Not that it's so important -- every camera shot will look different anyway from tape.

[sanlyn]

MCTemporalDenoise is slow. Very. It runs faster in MT mode, but I've never got MT to run on my machine because of all the older scripts and plugins I use. MCTD isn't absolutly necessary. Looks like DegrainMedian did OK in your own avi, it just doesn't clean up some other stuff -- but, then, running MCTD is like running 30 plugins at one time. The SmoothLevels() line is a very basic levels fix. Everything else was in VirtualDub.

[low-fat-al]

Thanks for that...

new error:

mt_lutxy : wrong colorspace, only YV12 and I420 allowed
(dither.avsi, line 1156)
(dither.avsi, line 1131)
(dither.avsi, line 303)

[sanlyn]

The mpg colorspace should already be YV12.

[poisondeathray]

If he's copy/pasting the script in post #18 the culprit is probably , Dither_convert_yuv_to_rgb

[poisondeathray]

Speed / housekeeping suggestion - It it appears you crop, addborders, then crop again ? Why not the just the 1st crop ?

[sanlyn]

Thick borders give filters like SmoothLevels a hard time--or, they give me a hard time fiddling with the code. They'll have to come off and/or be replaced later anyway, they won't be black after other corrections. I also temporarily remove them anyway when working on color: they're distracting.

Also: if I left the clip cropped, I had serious chroma disturbances with IVTC, and also with QTGMC earlier. Those two guys didn't like my crop, I guess.

[sanlyn]

If he's copy/pasting the script in post #18 the culprit is probably , Dither_convert_yuv_to_rgb

For some reason I found the dither package tricky to set up. Took me a while. I think Avisynth's ConvertToRGB32 ought to do it.

[sanlyn]

Okay, same script and all but tweaked levels/color for each camera shot. Tape is a pain in the neck. This one looks better, even if I left the audio a bit tinny again. Some of the blacks look a tad green, gamma's a pinch too high. Not quite there, but....better. Repeat: Tape source is a pain in the neck.