Problem NTSC & PAL interlaced source MPEG2

[TubeBar]

Nevermind, davexnet already posted.

OK I was able to load the scrip in VD and save video, it looks fine except the frame rate is wrong. MediaInfo is reporting it at 22.965 FPS (should be 23.976).

Any idea how to fix that besides just speeding it up but it sounds like it dropped more frames than it should have.

FFmpegSource2("S1E2 (NTSC) 29.97i.mpg")
TFM()
TDecimate()

Thanks again for all your help, going to get some rest be back later to try this agin.

[jagabo]

It looks like your source is variable frame rate. Change the ffmpegsourc2 line to:

FFVideoSource("S1E2 (NTSC) 29.97i.mpg", fpsnum=30000, fpsden=1001)

[poisondeathray]

Probably a source filter issue. ffms2 can be unreliable for mpeg2 sources. You can try rffmode=1 to honor pulldown flags. It works for some versions of ffms2. There are newer version of ffms2 in the doom9 thread

Most reliable source filter for DVD sources are DGIndex (MPEG2Source) , or DGIndexNV (DGSource ) if you have a compatible CUDA card


There are some scenes in that NTSC sample that are likely from a 25p content source. They are jerky at 23.976p . e.g the 1st scene pan, 2 guys in helicopter, pulling up to the door. You need a longer TDecimate cycle, cycleR window to recover all the film frames, especially for the 1st one. The srestored PAL version is jerky with dropped frames, dupllicated frames and blends. It's slightly better at 25p. Likely the PAL version was taken from the NTSC version which was taken from another PAL version, or at least parts of it were

Only the intro in the NTSC sample has chroma blends (main episode does not, or I didn't see any), but the intro looks 23.976p . It was probably processed separately. There are orphaned fields that can be used to "fix" the chroma blended frames, and/or you can replace with duplicates such as the YT video. If the intro is the same for all episodes, I'd probably fix one and use that for all.

The quality of the NTSC intro is significantly better overall than the PAL version in terms of fewer compression /artifacts, more details. The main episode is slightly better too

[poisondeathray]

Frame quality comparisons - the NTSC version is roughly matched in color to the PAL version (you could spend more time matching it more accurately, but the PAL version is technically wrong because of the elevated black level), and PAL version is cropped and scaled using Spline16 to align to NTSC frame

Rough color match NTSC to PAL version

Coloryuv(cont_u=12)
SmoothTweak(saturation=0.91)
SmoothLevels(0,1,250,15,255)

https://imgsli.com/MTk1OTk2/0/1

The first few are from the intro and show significant encoding artifacts in the PAL version. But those are from the intro. The PAL main episode doesn't seem as bad as the intro, but still worse than the NTSC version

The 2nd and 3rd last show are typical and illustrate that you'll always get some remaining blends with srestore and can often some bluriness from deinterlacing instead of field matching

The last one is perhaps the most revealing in the main episode - the PAL version wall textures, lines on the base of ship, shadow of the wall are all deteriorated.


Examine scenes with continuous animation. eg. The pan at beginning after the intro , when the ship is in the air with the pilot and passenger. Definitely not "23.976p". That pan is perfect cadence wise when you use proper IVTC settings (TDecimate with longer cycle, cycleR "window" - TDecimate(cycler=9, cycle=54)) back to "25p" - you can slow it down later with the audio +/- pitch correction if desired, but that gets you the original unique frames back. Anything else like 23.976p will have dropped and/or duplicated frames in those scenes. I would double check with a few more scenes with motion, but in that sample there was more than one, and it's certainly suggestive at least a few scenes were taken from a 25p source

I posted a side by side of the pan shot of the SRestored PAL version and the properly IVTCed NTSC version , slowed down to 1/4 speed. (Same thing in terms of color and scaling to match for the comparison)



Again, it depends what your goals are, but the NTSC version is better overall to start with. I would filter out the dotcrawl, rainbows, some minor noise, but those are subjective. Proper cadence, higher quality frames, lack of blends are preferred characteristics

[TubeBar]

It looks like your source is variable frame rate. Change the ffmpegsourc2 line to:

FFVideoSource("S1E2 (NTSC) 29.97i.mpg", fpsnum=30000, fpsden=1001)

Thanks that worked

Probably a source filter issue. ffms2 can be unreliable for mpeg2 sources. You can try rffmode=1 to honor pulldown flags. It works for some versions of ffms2. There are newer version of ffms2 in the doom9 thread

Most reliable source filter for DVD sources are DGIndex (MPEG2Source) , or DGIndexNV (DGSource ) if you have a compatible CUDA card


There are some scenes in that NTSC sample that are likely from a 25p content source. They are jerky at 23.976p . e.g the 1st scene pan, 2 guys in helicopter, pulling up to the door. You need a longer TDecimate cycle, cycleR window to recover all the film frames, especially for the 1st one. The srestored PAL version is jerky with dropped frames, dupllicated frames and blends. It's slightly better at 25p. Likely the PAL version was taken from the NTSC version which was taken from another PAL version, or at least parts of it were

Only the intro in the NTSC sample has chroma blends (main episode does not, or I didn't see any), but the intro looks 23.976p . It was probably processed separately. There are orphaned fields that can be used to "fix" the chroma blended frames, and/or you can replace with duplicates such as the YT video. If the intro is the same for all episodes, I'd probably fix one and use that for all.

The quality of the NTSC intro is significantly better overall than the PAL version in terms of fewer compression /artifacts, more details. The main episode is slightly better too

Yeah it's clearly different source w/ the intro I agree. Pretty bad DVD release unfortunately. It may require separate processing to get it perfect, not sure I can get it perfect. I'll fully review tomorrow and also take a look at what you suggested.

Thanks.

[TubeBar]

Frame quality comparisons - the NTSC version is roughly matched in color to the PAL version (you could spend more time matching it more accurately, but the PAL version is technically wrong because of the elevated black level), and PAL version is cropped and scaled using Spline16 to align to NTSC frame

Rough color match NTSC to PAL version

Coloryuv(cont_u=12)
SmoothTweak(saturation=0.91)
SmoothLevels(0,1,250,15,255)

https://imgsli.com/MTk1OTk2/0/1

The first few are from the intro and show significant encoding artifacts in the PAL version. But those are from the intro. The PAL main episode doesn't seem as bad as the intro, but still worse than the NTSC version

The 2nd and 3rd last show are typical and illustrate that you'll always get some remaining blends with srestore and can often some bluriness from deinterlacing instead of field matching

The last one is perhaps the most revealing in the main episode - the PAL version wall textures, lines on the base of ship, shadow of the wall are all deteriorated.


Examine scenes with continuous animation. eg. The pan at beginning after the intro , when the ship is in the air with the pilot and passenger. Definitely not "23.976p". That pan is perfect cadence wise when you use proper IVTC settings (TDecimate with longer cycle, cycleR "window" - TDecimate(cycler=9, cycle=54)) back to "25p" - you can slow it down later with the audio +/- pitch correction if desired, but that gets you the original unique frames back. Anything else like 23.976p will have dropped and/or duplicated frames in those scenes. I would double check with a few more scenes with motion, but in that sample there was more than one, and it's certainly suggestive at least a few scenes were taken from a 25p source

I posted a side by side of the pan shot of the SRestored PAL version and the properly IVTCed NTSC version , slowed down to 1/4 speed. (Same thing in terms of color and scaling to match for the comparison)



Again, it depends what your goals are, but the NTSC version is better overall to start with. I would filter out the dotcrawl, rainbows, some minor noise, but those are subjective. Proper cadence, higher quality frames, lack of blends are preferred characteristics

Sorry missed this when I posted the above. Thank you for the input, I will review tomorrow as tired now

I think perhaps the easiest way s to use the NTSC except for S1E1 the NTSC is totally botched Lower quality print, has PAL pitch for audio but it's slower than 23.976 (it's running much slower). I can upload a clip. So I have to use PAL for the 1st episode and I can use NTSC for the rest of the series. I think that makes more sense.

If you want a look at the full original VOBs (for E1 and E2) to see what I mean about the screw-up for E1 NTSC it's on my OneDrive: https://1drv.ms/f/s!AucfsqOdqyS9vkTejql1MasB1qjD

I can upload a small clip tomorrow if you don't want to deal w/ huge VOBs.

[TubeBar]

Frame quality comparisons - the NTSC version is roughly matched in color to the PAL version (you could spend more time matching it more accurately, but the PAL version is technically wrong because of the elevated black level), and PAL version is cropped and scaled using Spline16 to align to NTSC frame

Rough color match NTSC to PAL version

Coloryuv(cont_u=12)
SmoothTweak(saturation=0.91)
SmoothLevels(0,1,250,15,255)

https://imgsli.com/MTk1OTk2/0/1

Yes after carefully reviewing NTSC have better colors, PAL is a bit off. I like your corrections, not sure how to add that to script Lollo gave me? I attached his script.

The first few are from the intro and show significant encoding artifacts in the PAL version. But those are from the intro. The PAL main episode doesn't seem as bad as the intro, but still worse than the NTSC version

The 2nd and 3rd last show are typical and illustrate that you'll always get some remaining blends with srestore and can often some bluriness from deinterlacing instead of field matching

The last one is perhaps the most revealing in the main episode - the PAL version wall textures, lines on the base of ship, shadow of the wall are all deteriorated.

Examine scenes with continuous animation. eg. The pan at beginning after the intro , when the ship is in the air with the pilot and passenger. Definitely not "23.976p". That pan is perfect cadence wise when you use proper IVTC settings (TDecimate with longer cycle, cycleR "window" - TDecimate(cycler=9, cycle=54)) back to "25p" - you can slow it down later with the audio +/- pitch correction if desired, but that gets you the original unique frames back. Anything else like 23.976p will have dropped and/or duplicated frames in those scenes. I would double check with a few more scenes with motion, but in that sample there was more than one, and it's certainly suggestive at least a few scenes were taken from a 25p source

I posted a side by side of the pan shot of the SRestored PAL version and the properly IVTCed NTSC version , slowed down to 1/4 speed. (Same thing in terms of color and scaling to match for the comparison)

Again, it depends what your goals are, but the NTSC version is better overall to start with. I would filter out the dotcrawl, rainbows, some minor noise, but those are subjective. Proper cadence, higher quality frames, lack of blends are preferred characteristics

This is fun and has been a very interesting learning experience. I am not 100% convinced it's PAL sourced at 25 because of comparing the times to NTSC VHS captures on Archive.org. I was only able to find the first 2 episodes that were correctly captured I attached both of those.

At least that VHS capture (and others not done properly but they are NTSC) the speed, pitch and time match both PAL and NTSC DVDs indicating to me it's 23.976 sourced?

I think I will use NTSC for everything except the first episode. S1E1 NTSC DVD the pitch is wrong (to high) and quality of that film print looks like VHS and speed is wrong - it's running to slow almost 1 minute to long. Just not fixable. That's why for that epsiode as you can see I need to use PAL and if can figure out how to add your color correction should be OK.

I'm not entirely sure on speed issue? If keeping it at 25 is correct, going by the times from VHS properly captured clips.