Yet another interlacing issue

[Daodan]

I could pretty much take care of interlacing until now, but this problem really put me down. I have an interlaced PAL HD movie, frame based, bottom field first all the way. When I apply deinterlacing (tried mvbob, tdeint+eedi2, tdeint) I always get 50 fps (on some 29.97 fps interlaced clip I always get the same fps as input). I tried using selecteven, but some of the freames are ghosted badly (groups of frames at some interval of good frames). If I use selectodd, I get the other set of frames ghosted (not so badly though). Tried assume bff, tff and any other type of assume, etc. No luck.
I'll post a sample as soon as I can, but can someone give me for the moment any hints as to what should I do? Thank you.

[GodofaGap]

If you use bob-deinterlacing you will always get double-framerate output. So make sure you use a non-bobbing mode in your deinterlacer. So don't use e.g. mvbob or tdeint(mode=1).

[foxyshadis]

Ghosted on most of the frames, try Restore24. Sounds like it was blend-converted to 25.

[Daodan]

Tried that already. Horrible results.
By the way, even when playing the original file there is a lot of ghosting (when using deinterlacing in ffdshow, I tried all kinds of deinterlacings).

[foxyshadis]

Blending is one of those cruel and unusual issues that's hard to diagnose without a sample and even harder to fix with one.

It sounds like you might be victim of a simple overeager temporalsoften job, though. If so, not much you can do but try to minimize it and live with it, unless Didée or MOmonster have worked some magic behind my back.

[scharfis_brain]

provide an unprocessed sample clip

[Daodan]

Added adress for an unproccesed sample. Obviously still no improvements, I tried almost all deinterlacers out there last night.
http://www.savefile.com/projects/668353

[Kika]

That's an unprocessed Sample? Whoa! Looks like they upscaled a interlaced Movie whithout protecting the Fields.

[Daodan]

They did? Because it doesn't look DVD quality, it is better I would say. So anyway, is there any way to fix this? (yes, it's unprocessed).

[Kika]

Maybe they did a change in the AR, bad things like 2:35:1 to 1.85:1. What ever the source was, the resizing has destroyed the interlaced structure.

[scharfis_brain]

Kika, you are wrong!

The sample file contains 24 or 23.976 fps progressive film fieldblended into 50i.
The interlacing structure is fully intact!

theis script gives you a deblending without yerkiness:

mpeg2source("C:\Downloads\forum\daodan\[savefile]060315050932_sample_HD.d2v")#,cpu=6)
crop(0,0,0,-8,align=true)
tdeint(mode=1,type=3,tryweave=true, full=false)
changefps(24*10) #or 23.976*10
selectevery(10,x) replace x with any offset between 0...9. the sample needs 2

[Daodan]

Thanks a lot! I'll try this as soon as I get home. I almost gave up. By the way, is that cpu=6 really necessary? I know it's good for blend recognition, but if it's possible I would like to avoid blurring the image that much. Again thank you.

[scharfis_brain]

you know what the # char does, don't you?

[Daodan]

oups... didn't notice.

[Kika]

@scharfis_brain

Oops, you are right. Checked it by using mme. Darn, never trust a DirectShow-Filter (or don't work on videos in the office). :(
Sorry.

supplement: Now i installed VLC and get a correct picture, but still not in MPC. There must be something wrong with my filters. All i get is something like in the attachment.

[scharfis_brain]

@KiKa: set MPC's video size to "normal size" in "View -> Video Frame".

@Daodan: Restore24 *should* do a good job with this video, since it is an upconvert of 720p23.976 to 1080i50.

anyways you can use this script:
mpeg2source("C:\Downloads\forum\daodan\[savefile]060315050932_sample_HD.d2v")
crop(0,0,0,-8,align=true)
bilinearresize(1280,height)
#bob(height=720,0,1)
tdeint(mode=1,type=2,tryweave=true, full=false).lanczosresize(1280,720)
changefps(23.976*10.0)
selectevery(10,x)


As I wrote earlier, set x to an offset between 0...9 and scroll through your movie and watch for remained blended frames. Especially after scene cuts and/or advertisements.
If the blending starts getting visible again, you need to trim away the good portions from the bad portions and use another offset value.

the settings in the script and use for/of Tdeint are strongly recommended. For a big speed gain and also quality drop uncomment the bob(...) line and comment the tdeint(..)-line.

[Daodan]

The script you gave me earlier worked nicely, from what I looked through the whole stream. Restore24, as I said before makes a real mess, and I put in the slowest settings. I really wonder how they can broadcast such stuff since there is no method to play it real-time as it should.

[scharfis_brain]

the 2nd script has some speed improvement regarding resizing (as there is no reason to keep it 1080p).

Also I hope that you used the first script with 23.976 * 10 instead of 24 * 10.

I really don't get the meaning of your last sentence

[Daodan]

Yes, of course, resizing etc I was able to fit in myself. The last sentence is that this is the way they broadcasted the movie, how am I supposed to play it correctly on my computer?

[dukey]

Kika
it looks like CUE (chroma upsampling error)
i had a similar problem
Basically u wanna deinterlace it before it changes colour space or gets upscaled or anything
*i think*
its in the avisynth readme somewhere

[scharfis_brain]

ow am I supposed to play it correctly on my computer?enable bob-deinterlace in your mpeg-decoder or better wait until your encode has finished (the restored 23.976 fps video) and play that without any deinterlacer.

[Kika]

@dukey
No, it was an resizing problem, it's solved now. I had to reinstall some components of my system.

[Didée]

Just for the record ...

Restore24, as I said before makes a real mess, and I put in the slowest settings.
You must've made something wrong. Restore24 works perfectly on your sample.

I tried the most simple usage I can only think of:
Mpeg2Source("sample_HD.d2v")
Crop(0,0,-0-8)
LanczosResize(1280,height)
Restore24()
LanczosResize(1280,720)
Return(last)
and got the following

result of Restore24 (http://rapidshare.de/files/15752741/sample_HD__Daodan__R24-test1.avi.html). (4.6 MB, on Rapidshare.de)

Not one single error. 100% flawless result.

[Daodan]

I'm sorry to bring this up again but scharfis_brain's nice method had some side effects. I used an offset of 1 which eliminated the ghosting perfectly in all the movie, but I get some strange frames at very sudden transitions. The last frame from one scene and the first from the next look really bad, some color blocks and some other weird artefacts. Unfortunately I didn't see Didee's post (thanks!) so already encoded it, and overall it's not really noticeable those bad frames, but I would like to know what has caused that. It's some kind of bug in tdeint?

[Didée]

Probably not a bug, but perhaps some of the various parameters of TDeint could've been tweaked better, or at all.

Is the source still available, or already deleted?

[Daodan]

Still available. I tried to play with some parameters in Tdeint but no improvement. Which is the one that can cause this? probably if I reencode I'll use restore24, but I would also want to learn how to avoid things like that in the future.

[Didée]

Can't tell out of the sleeve - TDeint has *so* many options, and I'm really not *that* familiar with all of them.
One quick guess is that it's related to "tryweave=true" (which I'm often alittle suspicious about), and some thresholds should be lowered for that.
Another possibility, though not quite as likely, is that it's related to the often occuring "swinging back-and-forth" of the blending's pattern length (21-25-29 misery). It's currently not fully clear to me if scharfis_brain's stricly linear method occasionally can stumble over these, or if it's safe when the right offset is given.

In any case, a sample (original source) of one of those problematic scenes would help to evaluate the problem. Sample should start several seconds before the cruitial frames.

[scharfis_brain]

The strange blocking is already in the source.
It is caused due to the fieldblending and the low bitrate used to compress that 1080i video.

You cannot do much against it (You cannot polish a turd...)
You may use cpu=4 in the mpeg2source() command. That will smooth these blocks a little bit.

[Daodan]

Checked and triple checked. It's not in the source. (the source is quite blocky but what I'm talking about is extreme blockiness). I'll provide a sample as soon as I can.

[Daodan]

Uploaded a sample. Here:
http://s55.yousendit.com/d.aspx?id=1FXV750CADPJV0MV4NY7XNL43G
It has 3-4 scenes like that in it. Two of them when watched without deinterlacing (the source) seem to have some strange frames that don't appear with bob deinterlacing.