Log in

View Full Version : 29.97fps DivX to 25fps VCD


Boulder
5th March 2003, 20:08
Is there any way to do this frame rate conversion properly? I have a DivX file which has been encoded at 29.97fps and deinterlaced and I think that IVTC should have been done instead. If I simply use Decimate() with AssumeFPS(25), the result is jerky. Is that something I just have to live with or is there anything better that can be done?

scmccarthy
5th March 2003, 23:58
Try decimate(cycle=6)

If you IVTC it, you will have to speed up the audio with the video from 23.976 to 25.

Stephen

PS And unless the source was telecined, performing IVTC will make it look jerky. cycle=6 instead of cycle=5 reduces the number the frames that are dropped which should reduce the jerkiness somewhat. The only other alternative is to blend all the frames into a different frame rate. This will eliminate jerkiness and reduce sharpness instead.

DJ Bobo
6th March 2003, 01:16
@ scmccarthy
His source is already deinterlaced, this won't work.
I personally don't know a method to get smooth results (manual IVTC may be?!)

scmccarthy
6th March 2003, 06:06
@DJ Bobo
If it has been deinterlaced, the only thing left for a manual IVTC is to decimate the repeating fields. If there are none, then it was not originally telecined and therefore not a candidate for IVTC.

What else would a manual IVTC be?

decimate(cycle=5) gives
23.976fps which can be sped up to 25fps along with the audio.

decimate(cycle=6) gives
24.975fps which can be sped up to 25fps along with the audio:

audiodub(25,true)

Stephen

Boulder
6th March 2003, 13:58
Thanks for the advice. I didn't remember the old Decimate-by-six thingie so I'll experiment with that. I'm sure it will look better that way:)

DJ Bobo
6th March 2003, 16:07
@ scmccarthy
You don't get it, do you?
With deinterlaced, he certainly means blended (2 frames from 5 are blended). That means IVTC is not possible the way we know it.
Manual IVTC = finding the pattern manually, and discarding the redundant fields manually.
Which means he would use separatefields(), then use selectevery(...) to discard the redundant fields, then weave and eventually telecide and finally accelerate.

Using decimate (with any cycle) directly won't lead to any smooth results :p

Boulder
6th March 2003, 16:13
Whoa, lots of information for my tortured brain:scared:

I have never had to do anything to such source material so this is a new experience to me. I'll have to play with your suggestions and see what eventually comes out of it.

Boulder
6th March 2003, 16:40
I did some experiments and noticed that the original 29.97fps clip has two blurred frames in a row and then three sharper frames in a row. Can this be used to determine the necessary procedure and how should the conversion be done? Sorry for the simple questions but I'm a total newbie when it comes to this stuff.

DJ Bobo
6th March 2003, 17:00
Indeed, this is telecined video, but deinterlaced with blendfields.
AFAIK, it's impossible to get a correct conversion out of such material. I just noticed, even with separatefields(), you won't get rid of the blends :(

The best solution I can give you is:

separatefields()
selectevery(12,0,1,2,3,4,5,6,7,8,11)
weave()
blendfields()
assumefps(25)

This won't be smooth of course.

If anyone knows how to separate fields and get rid of the blends at the same time, please share with us.

Boulder
6th March 2003, 17:05
Well, anything that is better than simply decimating by five is good in this case. Luckily there's not much panning so the jerkiness is not that irritating. I'll try that script and see what it produces.

scmccarthy
6th March 2003, 21:53
@boulder

Do you need help making sure the audio stays in sync, or is that taking care of itself?

Stephen

Xesdeeni
6th March 2003, 22:22
@DJ BoboAFAIK, it's impossible to get a correct conversion out of such material.Many moons ago, while strolling through the Doctor Who Restoration Team Website (http://www.restoration-team.co.uk/), I came across something related to this.

In case you don't know, this site is home for a group of people trying to recover lost Dr. Who episodes. For a variety of reasons, including re-using old tapes to save money at the BBC and a fire at the BBC, many episodes were lost (or preserved only on black and white film or only as audio). However, these episodes were distributed for broadcast around the world as either copies, or more often, as a tape returned to the BBC and later destroyed or lost. Various people have stumbled on bits and pieces of these episodes, perhaps recorded on one of the first Beta VCRs, or perhaps a copy from a film camera pointed at at TV (I'm not making this up!).

Anyway, some of the recovered footage was from episodes converted from PAL to NTSC using the old four-field method, that ended up blending fields. If I recall correctly, one of the comments on the above site talks about a technique they developed to reverse this process to remove the blended frames and reclaim the original frames. I'm afraid I haven't been able to locate the quote myself, but I thought I'd let anyone interested know that there actually may be a way to reverse this blending, if the technique can be worked out.

Xesdeeni

Boulder
6th March 2003, 22:26
@Stephen: I did the audio frame rate conversion in the Avisynth script with the line "AssumeFPS(25.000,true)". I normally use BeSweet to do these conversions but it doesn't support 24.975->25fps.

@DJ Bobo: The result is very good considering that the source is not very good. I'm sure I can live with it, the jerkiness is quite hard to notice unless you start looking for it;)