PAL 25i to NTSC 30i proper conversion - HELP NEEDED

BACKGROUND INFORMATION

For my next preservation project, I'm working with a childhood classic: Thomas the Tank Engine & Friends. Thomas was shot on 35mm from 1984-2003 in 25fps, being that the production was based out of London. Once the third season began, the editor John L. Wright started to speed up footage for certain shots and would output the episodes in 25i for TV airings. This went on until 2008, when they switched to strictly progressive shooting for their new animation technique and have kept with it.

When Thomas was brought over to the US, Anchor Bay took up distribution. Their releases contained pulldown, but were plagued with dot crawl. Anchor Bay also only had access to lower end analog masters, unlike the UK that were given much better masters of the episodes.

Thomas hit DVD in the UK with multiple complete season releases. The DVDs are great quality, but in PAL/25i. The US has only one complete season release (dated back to 2004).

THE GOAL

Basically, I'm looking to convert the PAL DVD files to NTSC 30i so I can play it on all players in my household (many don't support 50hz material). I had dropped the file in Vegas and put it in an NTSC project. While I achieved my goal, the video has lots of horizontal linear shaking, especially when text was on the screen. There was no judder or any other problem, just the linear shaking.

I had also tried rendering it without any resampling, but when the interlaced footage appeared, it was very jittery.

I've seen many companies able to convert 25i material to 30i with no shaking, only ghosting due to pulldown, which is inevitable at this point, how did they do it? Is it even possible?

Here's a way in avisynth.

video=mpeg2source("clip.d2v")

audio=audio file goes here

audiodub(video,audio)

tfm()

assumefps("ntsc_film",sync_audio=true)

spline64resize(720,480)

limiter()

This will give you a 23.976 fps output. You would then use the encoder to implement a pulldown flag. This will probably show up as progressive on most modern hardware.  

EDIT: This is the basic method I'm using to backup all of my seinfeld pal discs to ntsc btw.

althor1138 said:

Here's a way in avisynth.

video=mpeg2source("clip.d2v")

audio=audio file goes here

audiodub(video,audio)

tfm()

assumefps("ntsc_film",sync_audio=true)

spline64resize(720,480)

limiter()

This will give you a 23.976 fps output. You would then use the encoder to implement a pulldown flag. This will probably show up as progressive on most modern hardware.  

EDIT: This is the basic method I'm using to backup all of my seinfeld pal discs to ntsc btw.

 I don't want a 23.976 output though. The show as shot in 25fps, and slowing it down would cause sync errors with the narration (synced at 25fps). I'm looking for 30i solutions, if at all possible.

This is a 30i solution.  If you speed up from 25fps to 30fps do you think it will be any better?  23.976fps with a pulldown to 29.97fps is the way to go. 

EDIT: btw this script will sync the audio to match the 23.976fps.  It's no more disturbing than the speedup applied to movies filmed in 24fps then converted to 25fps to fit on a pal disc.

This era of show can't be slowed down or sped up due to certain interlaced segments edited within the episode.

I need a direct conversion from 25 -> 30, while remaining interlaced. No speedup, only pulldown. The show was never intended to have its speed altered. The plethora of companies that have released the show on VHS/DVD from the late 80s till now have always used pulldown. It's a confusing scenario, but they got it done flawlessly somehow.

Here's a link that contains both progressive and interlaced material from the show, from the PAL DVD: https://drive.google.com/file/d/0B1-W6ByLSZVKV01jNThMN3J5YVU/view?usp=sharing

Hmm, I think I see what you mean.  Just to be certain, are you saying that an officially converted NTSC episode has exactly the same playing time as the original PAL episode?

Yup, every single one of them. If you didn't see the link above, it shows the PAL footage (also synced with the US dubs from the US DVDs). The footage contains both progressive footage and interlaced footage.

Ok I did some reading and I think this is what you are after:

mpeg2source("palclip.d2v")

bob(height=480)

bicubicresize(720,480)

changefps(60000,1001)

separatefields()

selectevery(4,0,3) #use selectevery(4,1,2) for odd field first.

Weave()

This is just copied from the wiki.

EDIT: Just tested this and it seems to do exactly what you want.  I fed it a pal clip and out popped a 30i clip with the exact same playing time.

althor1138 said:

Ok I did some reading and I think this is what you are after:

mpeg2source("palclip.d2v")

bob(height=480)

bicubicresize(720,480)

changefps(60000,1001)

separatefields()

selectevery(4,0,3) #use selectevery(4,1,2) for odd field first.

Weave()

This is just copied from the wiki.

EDIT: Just tested this and it seems to do exactly what you want.  I fed it a pal clip and out popped a 30i clip with the exact same playing time.

 Tried that, it does the strange horizontal linear shaking that every other program does. Maybe I'm asking too much? I've just seen professional DVDs of this show with stable video.

I'd go with:

mpeg2source("source.d2v")
loadCPlugin("yadif.dll")
yadif(mode=1) # Bob mode doubles framerate
BilinearResize(720,480) #bilinear is usually considered best for downscaling
ChangeFPS(59.94) # Creates duplicate fields. Use ConvertFPS if you prefer blended fields.
AssumeTFF()
SeparateFields()
SelectEvery(4,0,3)
Weave()
converttoyuy2(interlaced=true) #Assuming you are encoding with CCE

This is somewhat related so I thought i'd ask here.  I have some violin concerto's on laserdisc that were assumedly filmed in 25i because they are mostly all German.  They have however been converted to 29.97i for the laserdisc release. I don't see any blending just interlacing. 

My question is how could I get back to 25i? Basically, can I do this:

25i --> 29.97i --> 25i  ?

Here is a 100 frame sample. I didn't see any blending but I could be wrong.

EDIT: I don't know. I believe there is blending going on.  I'm just not that familiar enough to say yes for sure.

You could also try RePAL().  It's designed specifically for 25i to 29.97i conversions.

Agreed, the quality of the bobber is the most important.  I do recommend testing a couple different ones.  Sometimes the slowest/most powerful doesn't produce the best results when blended fields are present, but mostly the slower the better.

To explain, bobbers separate the fields into 2 frames (odd field and even field), adjust the height so they line up, and then upscale to full resolution.

Smartbobbers work similarly, but instead of a dumb resize, they extrapolate the missing data from the other field and surrounding frames.

My thoughts on smartbobbers:

QTGMC is the best.  Try the default 'Slower' setting first.  There are faster settings that don't always look that different, but there is a point when it's better to just go to something else because the results are only slightly less slow but crappy looking.

YadifMod, supposedly an improved version of Yadif, but I can't remember actually ever using it.  I usually test it on my videos and then go with something else.

Yadif: If your computer is more than 3-5 years old, this is probably ideal.  It's fast and produces really impressive results considering.

There are other smartbobbers out there, but this is my toolbox.

Thanks for the tips guys.  I've now tried both methods and repal seems to do the best in this situation but it's not perfect.

After some deliberation, I've decided to just leave it as it is.  I'm going to be encoding these and putting them on blu ray so I think leaving them in an NTSC format will be best for compatibility and it doesn't look bad during playback to me. If the fields weren't blended I'd probably invest the time into converting them back to PAL.

Btw, I noticed a couple of scenes with really high motion that QTGMC couldn't handle with default settings.  I changed search to search=5 and searchparam to searchparam=32 and it worked just fine so it seems the default motion analysis settings for mvtools isn't quite robust enough to catch really high motion. You can get search=5 if you use the placebo preset but searchparam never goes over 2 unless you set it manually.  To me this means if something moves more than 2 blocks in between frames, then it is not processed properly and you get ghosting or blurring.  Maybe I'm misunderstanding the documentation. Otherwise, qtgmc is a really great tool imo.