NTSC to PAL - Need quality software conversion.

[Poppa_Meth]

I'm helping out with a little project that is going to require conversion of NTSC material into what will eventually be a PAL DVD release. The source I can get from the original in pretty much any format needed (it's output from CG animation.) What I need is a program that can convert that material into PAL format. I've read several guides on it but most free methods seem to have their share of problems, so I'm looking for perhaps a commercial piece of software that can do a better quality job on this. Price isn't a big factor at this point (that may change after seeing the price ) so give me links to anything you've know that can do a proffesional job. Also I have to get with the actual author on this but I'm fairly certain the source is output at 29.97FPS instead of FILM. Any help is appreciated.

[guns1inger]

There isn't a good software solution to format conversion. What you have read pretty much covers it. The DGPulldown methods seem to be the best, but you will still have to re-encode to PAL resolution. Results are best if you can get 23.976 output to work with. Certainly, nothing that involves actually adding or subtracting frames from the sequence produces anything but choppy, jerky results.

[mikiem]

From what little I know, color spaces etc shouldn't be as big an issue if it's anime or animated etc... FPS might not be as big an issue either if you can get (or get to) original, *IF* it was rendered at 12 - 15 then doubled -- that sort of thing. V/Dub would probably be fastest doing frame cropping/resizing. Sub streams could be a big pain, & there I'm including button highlights. The whole menu part get's nasty IMHO if you try to convert rather then recreate. Resizing anything with text causes enough distortion I'd even consider OCRing the subs to text.

"I'm fairly certain the source is output at 29.97FPS instead of FILM"

If it was frame doubled, not a huge prob., if telecined maybe try and find out info re: method for ivt, but if rendered directly to 29.976, you really have nothing to remove -- might be only choice to re-render in something like Vegas, creating all new frames from existing footage (like taking 25 stills per second whilst it's playing) which is a hefty render hit which will lose quality. In fact, might work best if you did just that, do an analog capture of NTSC at PAL specs.

[Safesurfer]

Couple of software solutions I've heard of:-

http://www.puremotion.com/dvconverter/index.htm

http://www.focusinfo.com/solutions/catalog.asp?id=8

[jagabo]

If you can get the source in any format you want have it rendered at 720x576, 25 interlaced frames per second, MPEG2 with DVD specs. Then all you have to do is master and burn.

[Poppa_Meth]

I appreciate the input. I still haven't heard back from him about the info I need. I'm hoping however he did it, he'll be able to encode directly to 720 X 576 at 25fps, but I'm not sure what output is available to him. I know he used Poser for part of it and another popular CG program (name escapes me at the moment) for the rest.

[W_Eagle]

Best standard conversion software is Canopus Procoder 1.xx.

[mikiem]

Did some reading yesterday -- had a bunch of docs from SnellWilcox that I'd been neglecting for ages. One doc was on std.s conversion...

My suggestion if you have to deal with pure NTSC, would be to rent a box, the same way a post house would rent a vtr in a particular format.

Straight Resampling (capturing a video stream) in another std will apparently work, but you get some quality loss from the capture process, as with any capture. I was thinking it might work better if you used an up-converting DVD player, but then you'd have to find something that would capture it, let alone in another std.

According to the doc, motion analysis & prediction (same basic idea as used in mpg2 encoding) is the best method for creating (interpolating) frames -- makes sense. When/if you have 24 p material & can remove 3:2 pulldown, you can get 24 discreet frames per/sec and can then convert to 25 as usual, but if you can't get to those frames (if they never existed), you have to take (create) a new series of snapshots at 25 fps.

Prob. is that *without* motion prediction you wind up with morphed frames -- a composite of one frame overlaid on another, with objects appearing twice (like a NTSC telecine added frame). Vegas I think is able to take another approach, a fairly wasteful one , of adding a bunch of intermediate frames between the originals at 29.976. Assuming there were enough of these intermediates created, saving only those frames at the 25 fps interval can be a bit more accurate, but you still get some reduced morphing -- you get more of a weighted average, which is less noticable then trying to create a composite using just 2 frames. Motion prediction would try to determine the movement in the frame, and then create a new one without laying one frame on another.

Another possibility, if your orig was at 29.976 without any pulldown added, might be to cut the number of frames in half, but only since it's animation... Traditionally animation was created at 12 - 15 fps, then padded to NTSC &/or PAL, and by halving the number of frames 1st, *think* the overall motion might be better preserved. Wouldn't need anything special in the way of software or equipment, but I don't think the results would equal those of a box designed for std.s conversion.

Of course it could be done is software written for that purpose with motion prediction built-in, but I don't think you'd be able to rent software used by a broadcast facility as you would a piece of hardware.

[Edmund Blackadder]

The absolute best software for PAL<->NTSC conversion (yes, even better than ProCoder or Vegas) that I've ever tried is HiCon32. It's the closest result to Snell & Wilcox hardware you'll ever get at the present moment in software-only solution. It uses a really good motion estimation technique. The horizontal fields lines are not dancing, the sharpness is as good as it gets and the motion, on reasonably fast moving images, is as smooth as on the original (no blurring or jerkiness). It's not quite as good as Snell & Wilcox on very difficult shaky camera footage, but it's the best I've seen (for a software solution). It's also one of the best software deinterlacers I've tried.

However, unfortunately HiCon32 is very slow and has a lot of bugs and lack some features. Try not to convert more than 9 seconds at a time because of errors if you try to convert more at a time (prepare a sequence of little clips out of the main clip to convert them without errors and later reconnect in NLE). It does not open the sound, so you'll have to synch it to the converted clips later in NLE. Also, the only good quality option for output codec is Uncompressed. Everything else is not worth using for serious results (most of the other output codecs for some reason are unseen by HiCon32).

But even though there are several major shortcomings, with a lot of time and patience the results are simply outstanding and make even ProCoder and Vegas look mediocre in comparison.

Here's the link to a trial download:

http://www.hhi.fraunhofer.de/german/bs/hicon/index.html

See their sample images too, what they show is pretty much what you're gonna get.

[Poppa_Meth]

We'll have to look into renting some hardware or perhaps take a look at HiCon32 though doing an almost 2 hour movie in 9 second clips may be more than I can stomach. I still haven't heard back from him on what he is capable of outputing. All this could be for nothing but I may as well find out while I can. Also we are both in the US so neither of us really has access to any good PAL equipment, ruling out the playback and capture on another machine option.

[Edmund Blackadder]

We'll have to look into renting some hardware or perhaps take a look at HiCon32 though doing an almost 2 hour movie in 9 second clips may be more than I can stomach.

Perhaps if you tell Fraunhofer HHI that you'll only consider purchasing the software if they fix the 9 second bugs, among with at least including the the lossless codec output option, maybe then they'll be motivated to release a whole new updated version .

It's really a shame that such a great motion estimation code is enclosed into such a buggy and underfeatured capsule.

P.S.: I tried to do a 2 minute PAL to NTSC test with HiCon32 some time ago in 9 second pieces, and that was about all I could take before going nuts. But the end result was absolutely great .