Detelecine / decomb issues

[Sosigene]

Hi,

I am new to the forum.

I am trying to convert my wife "Friends" DVD collection to mkv (H.264) so that she will be able to watch them on our media center. I started by ripping the entire 10 seasons to .iso and then .vob.
And now, I am trying to convert these VOB to mkv with handbrake. Since this TV show was telecine in 29.976 (2 interlaced frames followed by 3 progressives), I tried different options in handbrake to achieve a proper "detelecine".

However, with the "detelecine" option and "same as source", I get some slowdown in some part of an episode, not necessary when there is a lot of motion...
The "Decomb" filter is a bit better, but there is still some noticeable slow down.
I then tried to force the output to 23.976 (I tried Peak and constant). It is a bit better but also introduces jerkiness in some panning.


I read everything (*everything*) that I could find on this forum and on Google.
At this point, I really don't care to use another tool than handbrake to do a good detelecine.

Could you help me to figure out what I am doing wrong?

[edDV]

Detelecine in Default mode should work if the source is interlace (telecine). Are you sure?


Is this disc from TV caps or a commercial disc? I'd expect a Friends DVD to be mastered progressive with pulldown flags. I don't have one to check.

Do you see split fields when you frame advance the source in VLC or Virtualdub?

[Sosigene]

Hi edDV,

I am glad that you answered that quickly, thank you. I already read about everything you wrote about detelecine on this forum, so your name is not a new one for me

Yes, I am sure that my source is telecine. I manually checked the vob file with VLC frame by frame. I clearly have a "2 interlaces - 3 progressives" pattern. By "split field" I guess you mean an interlaced frame where you can visibly see "combing", right? In that case, yes I can see them.

I have all the original commercial discs, so this is neither a TV Cap nor something I downloaded on the net. Furthermore, it is not the cheap "collector" set they released a couple of years ago (don't know if it was mastered differently)

Could you confirm that by "mastered progressive with pulldown flags" you mean a SOFT telecine? If so, it is not the case since with soft telecine I would not see the 2-3 pattern when read frame by frame.
I think that the studio was a bit lazy when they released those DVDs.

I tried both Detelecine and Decomb, one at a time, with default settings. They are doing their job correctly since there are no more interlaced frames on the result. However, I have theses slow-down that are very weird. Detelecine is worse than decomb on this point.

thanks!

[edDV]

I clearly have a "2 interlaces - 3 progressives" pattern. By "split field" I guess you mean an interlaced frame where you can visibly see "combing", right?

Right and the combing will show fields from different frames. This shows most in motion scenes.

Could you confirm that by "mastered progressive with pulldown flags" you mean a SOFT telecine?

Yes. The Friends series was shot progressive* so a progressive DVD would have been the way to go, but interlace would be cheaper to produce. This could be because they were holding progressive for the Blu-Ray release.

I tried both Detelecine and Decomb, one at a time, with default settings. They are doing their job correctly since there are no more interlaced frames on the result. However, I have theses slow-down that are very weird. Detelecine is worse than decomb on this point.

I've never seen this "slow down" using Handbrake so can't respond. It may be player related. Maybe others have seen this.


* Multi-cam Panovision 35mm film + later years to HDCAM digital. The original edits would have been done interlace (telecine).

[Sosigene]

This is an example of combing in my file. This will be followed by another similar interlaced frame and then 3 progressives:



I will post a extract from my output to show you the slow-down...

[Sosigene]

oh and I forgot to tell you that my player is VLC, so this slow down is not related to some hardware

[jagabo]

Post before and after video samples.

I don't really use Handbrake but I seem to recall the (de)interlace options were confusing.

[Sosigene]

ok, I am working on that..

[edDV]

Yes this cap shows fields from different frames.

[Sosigene]

Ok, here my samples.

I did not send you the original mpg2 sample as it would have been too big. Instead, I attached a file where the mkv was created with no filter (OFF). Let me know if it is good enough.

You will also find a file where I applied the default decomb filter and one with the default detelecine.

You will see the slow-down I am talking about when the scene changes and Ross turns his body. This is one of the many examples I have in a single episode.

[jagabo]

Detelecine.mkv is encoded with a variable frame rate (though it's nominally 23.976) and has the frame rate flag set to 29.97. That may be confusing the player you're using. Every frame is unique and there are no comb artifacts except for the first frame (not unusual).

Decomb.mkv is encoded with variable frame rate (though it's nominally 29.97) and is flagged as 29.97. Every 5th frame is a duplicate. It plays back with six little jerks every second.

Off.mkv is variable frame rate (nominally 29.97) and flagged 29.97. The frames are still interlaced but are encoded as if they were progressive. A player might still show it without comb artifacts if it is set to deinterlace all the time. Or if it's played back through a composite or s-video cable (which is always interlaced) it may look fine on TV if the field order (TFF) happens to be correct for the device putting out the signal.

I did a quick test with Handbrake and hard telecined clip. I used Detelecine Default, and manually set the frame rate to 23.976. That worked properly. Sample attached.

[Sosigene]

23.976 with peak frame rate or constant frame rate?

[Sosigene]

Off.mkv is variable frame rate (nominally 29.97) and flagged 29.97. The frames are still interlaced but are encoded as if they were progressive. A player might still show it without comb artifacts if it is set to deinterlace all the time. Or if it's played back through a composite or s-video cable (which is always interlaced) it may look fine on TV if the field order (TFF) happens to be correct for the device putting out the signal.

This is because handbrake is still not able to set the interlace flag correctly in its output, right?

[jagabo]

I added a sample above. I didn't tick the Peak Framerate (VFR) option. You want constant frame rate. As far as I know, there's no way to get Handbrake to encode interlaced. On the advanced tab it looks like you can type in extra options but it doesn't appear to pass them to the encoder.

[Sosigene]

What about this sample? It is from a different scene.

my detelecine has been done with default parameters and a constant frame rate of 23.976 as you suggested.
However, the weird slow down or sluggishness has still be introduced.

You can see it in the scene where Monica is full screen and seems surprised.

Is there anything better than handbrake to do this kind of job?

[jagabo]

Yes, there are several duplicate frames in the CRF version. In the unfiltered VFR version there is an odd cadence in the telecine pattern. I suspect the "problem" is in the DVD -- maybe they slowed down or sped up the video for effect. Can you post that same section non-reencoded? You can use DgIndex or Mpg2Cut2 to mark a segment and save it as M2V or MPG.

[Sosigene]

sure... hold on.

and thank you very much by the way

[Sosigene]

Here is the original mpeg2 sample for the Monica scene. I used Mpg2Cut2.

[jagabo]

Yes, the telecine pattern is different at the start of that shot, even in the original MPG. The video was sped up by 25 percent. No doubt they wanted her to look more surprised by speeding up her reaction. Worse, it was done in such a way that there is only one field from one film frame, then the next film frame is repeated for 3 fields. Ie, instead of 2:3 pulldown pattern it uses a 1:3 pulldown pattern. That is confusing the Detelecine algorithm and causing the duplicate frames.

You could manually fix that every time it happens but it would be a big pain.

[Sosigene]

humm... I just hope they finally release those HD versions so that I can put the DVDs in a box never to been seen again

Anyway, in the meantime, what can I do with that?

Keep it as is? Handbrake doesn't support it very well (interlace flag)
or maybe deinterlacing (with slow) every frames while keeping same as source? I would lost quality with this scenario, isn't it?

You could manually fix that every time it happens but it would be a big pain.

Ten seasons, about 24 episodes per season... no way

[Selur]

just wondering if you analyse the original vob with mediainfo does it also identify the source as:

Scan type : Interlaced
Scan type : Interlaced
Scan order : TFF
Scan order : Top Field First
Interlacement : TFF
Interlacement : Top Field First

or did Mpg2Cut MessUp the flags?

[edDV]

humm... I just hope they finally release those HD versions so that I can put the DVDs in a box never to been seen again

Anyway, in the meantime, what can I do with that?

Keep it as is? Handbrake doesn't support it very well (interlace flag)
or maybe deinterlacing (with slow) every frames while keeping same as source? I would lost quality with this scenario, isn't it?

You could manually fix that every time it happens but it would be a big pain.

Ten seasons, about 24 episodes per season... no way

Best to keep them interlace and let the HDTV deal with it*. But Handbrake does not support interlace encoding. Choices are detelecine, decomb, deinterlace or a simple weave (all off) to progressive h.264. The x.264 encoder is capable of interlace encoding but most popular GIUs don't offer that option.

When I need to encode to interlace h.264, I use Vegas' Sony AVC codec.

Another option is to keep them MPeg2.


* current HDTV processors will either attempt inverse telecine or simply upscale fields to 59.94 at screen resolution.

[jagabo]

My usual process consists of:

1) Rip DVD with DVD Fab.
2) Use DgIndex to build an index file and demux the audio.
3) Use an AviSynth script to open the index file and perform any filtering.
4) Use x264 CLI to encode.
5) Mux x264 video and AC3 audio (from step 2) with mmg.

Using that method you could leave the video interlaced and encode interlaced. Or you could use a smart bob doubler (like Yadif or QTGMC) and encode 59.94 fps progressive. Either of those would be as smooth as the MPG source. Unless the player has problems playing the resulting file. Attached is a sample encoded interlaced with the above process.

[jagabo]

Ooh, I was able to get Handbrake to encode interlaced! On the Advanced tab add "tff" in the CLI options box at the bottom of the frame. It's case sensitive and you don't get any warning if the case is wrong. If there are options there already be sure to put a colon between the other options and tff. For example:

otheroptions=whatever:tff

So leave all the deinterlacing options off, add the tff flag, and encode. MediaInfo shows the scan type to be MBAFF and you can see the tff flag in the x264 command line options. Then hope your players play it properly.

<edit>
I just wanted to add, that you should use bff instead of tff when your source is bottom field first.

otheroptions=whatever:bff

The interlace options didn't work when I tried it in the past because I used caps instead of lower case!

</edit>

[Sosigene]

@Selur:

Just checked and original VOB is the same:

Scan type : Interlaced
Scan order : Top Field First

[Sosigene]

ok, I will try the handbrake interlaced option found by Jagabo first. Seems to be more straightforward and I will easily be able to compare it to the sample he provided from his own process.

I will also have a look to Jagabo process (using Avisynth and DgIndex) and edTV proposition (Vegas' Sony AVC codec, which I guess is not free. I have to check)

In the end, I want something that will play nice on:
- VLC
- Ipad using a mkv player
- XBMC
- WD TV box.

I'll test your propositions and let you know my results as soon as possible.

Thank you very much Jagabo and edTV for your help. Very much appreciated.

[jagabo]

@Selur:

Just checked and original VOB is the same:

Scan type : Interlaced
Scan order : Top Field First

Yes, I've never seen Mpg2Cut2 screw up. It has no ability to reencode. And the MPG file is indeed interlaced, TFF.

[Sosigene]

Just tried Handbrake with tff option, same as source VFR...

When played it in VLC with deinterlaced set to AUTO and mode to Yadif, the result is very good, I don't have slow down anymore.
The deinterlaced flag is correctly read by VLC,

I wonder why there is no documentation about this... It would be useful for a lot of people.

[jagabo]

I wonder why there is no documentation about this... It would be useful for a lot of people.

All i've seen is: http://trac.handbrake.fr/wiki/x264Options#howto

[edDV]

and edTV proposition (Vegas' Sony AVC codec, which I guess is not free. I have to check)

The Sony AVC codec comes with Vegas Platinum or Pro editing software (not free).