inverse telecine advice please

[perdomot]

I'm converting some vhs tapes that have anime on them to vcd and understand that doing an inverse telecine will increase the quality of the video when I convert the avi to mpeg. Unfortunately I haven't found any how-to guides to tell me how to do this. I'm capturing with VDub using the Huffyuv codec at 29.97fps and using TMPEG to encode. Can someone tell me how to use the IVTC in Vdub and whether I choose NTSC or NTSC Film in TMPEG when I encode? Thanks folks.

[lpn1160]

ivtc can be a little tricky. 1st off was the original vid telecined? if not then you'll just get jerky motion. in tmpg one should choose "film". as for my self i use avisynth when my original source is telecined. the script reads like this
Telecide()
Decimate(5)
or go here and read about ivtc

http://www.animemusicvideos.org/guides/avtech/avsivtc.html

one should also make a test to see if your player can play them back without jerky motion
sometimes it better to just stick with 29.970fps, the to go the 23.976fps route. really read up on it
good luck!!!

[DJRumpy]

Open your AVI in VirtualDub. Use the DoubleArrow button at the bottom of the VDub window, to step through your video, frame by frame. The frame number should only be incrementing by 1 each click. If it's incrementing by more, then your clicking the wrong button.

Examine your video. Try to find a high motion section (move the slider to find a good spot). You should notice a pattern of 3 progressive frames, and 2 interlaced frames. That is the patter for a telecined video. If every frame looks interlaced, then you should not perform IVTC.

You can find more info with examples of interlaced frames here:
www.lukesvideo.com

[perdomot]

Thanks for the help. I'm trying to use avisynth but its telling me that it cannot determine the file type when I try to open up the script. I followed the instructions exactly so I don't know what's going on. Any ideas?

[perdomot]

OK, it looks like my video is telecided so this is what I'm going to try. Please tell me if this is correct:
I'm going to use VDub with the telecide filter and then begin frameserving with VDub.
Then I'm going to use TMPEG with the NTSC format, not the NTSC Film, to encode the avi to mpeg.

Will this convert the video from 29.97 to 23.97 properly? If not, can someone tell me what I need to do? Thanks.

[Bondiablo]

unless there's another reason why you're frameserving from virtualdub it would be easier to just load your avi in TMPGEnc and start with one of the NTSC Film templates. this will set your frame rate and inverse telecine will already be checked on the advanced tab.

that's what i did when i tried converting animation and noticed a sort of flicker when there was motion on screen equivalent to a camera panning across the scene. the solution was either encode to NTSC Film and use inverse telecine or just leave it interlaced.

[perdomot]

Well, the way I understand it, the IVTC filter in TMPEG isn't very good compared to others like avisynth and VDub. Something about the way it handles it causes flickers according to some other posters. How was the quality of your mpeg when you compared the non IVTC with the IVTC version and could you see a difference on your TV? Thanks.

[Bondiablo]

I was getting a flicker before I used inverse telecine. After, the video looked great. Here's a clip if you want to see for yourself. Keep in mind this is encoded to standard VCD settings 1150CBR, etc..

http://home.att.net/~bonfiles/gargtestF.mpg

[perdomot]

Looks good.
What procedure did you use to capture and convert? I like to use VDub with the Huffyuv codec to capture and have been using the NTSC profile in TMPEG wizard to encode. Should I use the NTSC Film template instead? My main concern is with minimizing artifacts and blockiness. I just finished encoding with the VDub IVTC filter and the TMPEG filter and I'm going to burn it to disc to see if I can spot a difference on my TV. Do you know what the main benefits of inverse telecine are supposed to be so I can watch for them? Thanks.

[lpn1160]

(STORAGE) one benefit is that, @1150 kbps with 23.976fps there is a little more data that can be allocated than @ 29.970fps. and one also gets the full progressive frames that the original source was filmed in.
(short and simple answer)

[DJRumpy]

If your using TMPGenc, then the FILM option is the correct one. After you IVTC a video, it's essentially been converted back to FILM (23.976fps).

I've never tried to filter in TMPGenc. I'm not happy with the current crop of IVTC filters. They always seem to guess the pattern wrong. During any video, the pattern of 3:2:3:2 can often change. A video may have a normal progressive stream, and then jump into an interlaced scene. This usually happens becuase mutliple people are working on the video/movie. There are scene splices, cuts, fades, etc. Adaptive filters were thought up to handle these changes in the patten. These adaptive filters try to spot the duplicate frames in a movie, but video noise can make the filter guess wrong. If it guesses wrong, and removes a frame, that was not a duplicate, your video will 'jump'.

If anyone has a good filter for either AVISynth, or VirtualDub, that doesn't guess wrong half the time, I'd be interested.

I've used the filters with MPEG2DEC, and the Frame Rate filter in VirtualDub ([Source] no change...[Frame Rate Decimation] - None...[Inverse Telecine] - Adaptive) to IVTC. Any more, I just leave it telecined, since the output is for television anyway. Plenty of room on a DVD...

If you do suggest a good IVTC filter, please include the settings used.

[batman0105]

DJRumpy:

I couldn't agree with you more about the current selection of IVTC filters. I find they are especially troublesome to work with when your source is a noisy VHS cap, specifically dark scenes. However, I have found that the GreedyHMA Avisynth filter does a decent job, not perfect mind you, but decent.

Here is the setting I have been using with some success:
GreedyHMA(1,0,4,0,1,0,0,0)

I don't remember what each setting signifies off the top of my head, but the documentation is pretty clear about explaining them. You can download this filter at www.avisynth.org.

Another method I am experimenting with that I hope will help increase the effectiveness of the IVTC process is to run a temporal filter before GreedyHMA.

Good luck!

[menace]

gentlemen--rather than start yet another thread on inverse telecining I thought I would attach my question to this one--

Does anyone know of a good guide that walks yuo through Tmpgenc's inverse telecine process--i.e. settings, fixing bad frames where the prog guessed wrong etc.? It seems like it might be a powerful utility if you knew how to use it. Thx!

[hustbaer]

hi!

@DJRumpy:

i thinkt you'll already know that one, but i'm very happy with the "Decomb" filter package for avisynth 2.07.
i haven't yet tried avisynth 2.5 since it's beta and i hate being beta-tester when i want to get work done.

however, if you have not tested it yet, give it a try.
i use standard settings except "post=true" for the Telecine() filter.

decomb does NOT use a fixed IVTC pattern, and it can even be used to IVTC telecined PAL->NTSC or FILM->PAL material (FILM->NTSC of course works too, it's the default).
you can activate the "pattern guidance" mode if the source is very noisy, so decomb will only break a patten once found after at least N mismatches, where N can be set anywhere from 1 up to 100.

a help file explaining all settings is included in the .zip file.

if you try this one, please let me know how you liked the results.

links:

http://avisynth2.sourceforge.net/ (avisynth project homepage)
http://shelob.mordor.net/dgraft/mine.html (decomb's author's homepage)

bye,
--hustbaer

[hustbaer]

@batman0105:

of course if you want to give it a try... do it.
but be warned:

IVTC filters should be first in chain.
maybe (!) a spatial-blur could help, but the temporal-blur (or temporal-soften) just makes it harder for the IVTC filter to guess the pulldown pattern, since it will "blend-in" information from the neighboring frames, hence changing the "duplicated" field, hence hindering the IVTC in "seeing" that it's a copy.

so i'd suggest:
IVTC -> spatial-soften -> resize -> crop

if using chroma-denoise (good for VHS) plug it in directly after IVTC, or maybe before.

whatsoever, good luck,
may the force always be with you,
bye,
--hustbaer

[perdomot]

Is there a difference between spatial soften and temporal soften? I'm using the Peach temporal soften for my anime and it seems to work fine. Mostly, I want to try and minimize artifacts and other junk when I encode to mpeg. Ive been letting TMPEG do my resizing and cropping for me but do you think I should include that in my script or does it make any real difference? Thanks.

[DJRumpy]

Spatial Soften filters typically work in the same frame in a radius around the reference pixel. Temporal Filters tend to work across multiple frames, so if a bit of noise is on 1 frame, and not on the adjacent frame, and it falls within the filters chroma range, it will remove it. I find that temporal filters usually do a very good job with analog noise (snow). Spatial Filters tend to do a better job with pixel noise (like macroblocking), at the cost of contrast. They tend to blur your color ranges into a more uniform color. For example, a red to orange blend from a rainbow, might be broken into multiple different distinct color bands by a Spatial filter, because it blends simular chroma values together.

This has the end result of making people look somewhat 'plastic', as the slight shading in real life is removed by the filter.

hustbaer, you don't necessarily have to put the IVTC filter first. If you capturing from VHS, you can crop without fear of fudging the filter's attempts at IVTC. The noisy video edges from VHS can interfere with a filter as well (or any video source with noisy edges), so it's a good idea to crop them off, or black them out before running IVTC. As long as you are only cropping you should be fine, because, as you already noted, any other filters will tend to make it more difficult for the IVTC filter to identify the dupilcate fields.

[batman0105]

Hustbaer:

Thanks for the reply!

I did a test running FluxSmooth() before GreedyHMA on a particularly noisy clip that previously had been quite jumpy after IVTC and I found it was noticeably smoother, but I think I just got lucky in this instance as I am inclined to agree with you that running such a filter before an IVTC filter is generally not a good idea.

However, with regard to the Decomb filter, I am curious to know what pattern guidance settings you have had success with. I have used Decomb in the past and for DVD sources it works great, but for noisy VHS it seemed to produce poorer results, with frequent jumpiness, especially in dark scenes where GreedyHMA was smooth.

[hustbaer]

@DJRumpy:

ok, right, you can crop before IVTC.
just make sure you always crop even amounts of lines from the top.
otherwise you might change the field-order.

any comments on the decomb() package?
already know that one? like it or don't?

@perdomot:

cartoons, or "animated-films" are a little bit different in every way.
IVTC don't really work very well with them, because they are usually done with 8-12fps, but not necesarrily - some change the pattern frequently, and so on.
i've never tried to convert cartoons and there sure is no one good solution for all different kinds of cartoons, so whatever works for you will sure be fine!
temporal soften filters work fine for normal movies too, but i'd always put them behind the IVTC filter.
that's what i ment to say.

bye,
--hustbaer

[DJRumpy]

Good point about the cropping. You could, in essence, change a top field, to a bottom field, and vice versa. I'll double check to see which IVTC dll I'm using. It's for version 2.0 of AVISynth (legacyIVTC.dll, or somesuch). I'll have to check when I get home.