Telecide producing very small mismatches

[k0r0n4]

I'm trying to tweak Telecide to grab some very small field mismatches in some video I'm IVTCing. 99.99% of the IVTC is excellent thanks to the filter, but there are several frames that aren't. I'll let the images speak for themselves.

http://img.photobucket.com/albums/v16/yoitsmeremember/Doom9/bad-ivtc.jpg

Don't see it? Take a closer look at her mouth:
http://img.photobucket.com/albums/v16/yoitsmeremember/Doom9/bad-ivtc_zoom.gif

Now, I know there is no perfect IVTC, and this may just be one of those cases where it cannot be fixed. I thought of a few other workarounds to solving this problem. The first being using deflicker with a window of 1 to correct it (VERY scary, as her mouth would probably never open or close with that set), and the second being tweaking the settings somehow on Telecide. Here's the line I'm currently using.
Telecide(order=1,guide=0,gthresh=10,post=1,vthresh=50,dthresh=7,blend=false,show=true,chroma=true,back=2,bthresh=50,nt=10,y0=1,y1=1,hints=true,ovr="",debug=false)
Obviously, the guide was the first thing I tried adjusting. Setting it to 1 didn't corrent the bad match at all....in fact, it made more errors on the mouth on other frames. The source is NOT true interlaced and NOT hybrid either.

Any ideas on how to fix this, or do I just have to live with the imperfections of automatic IVTC?

[hartford]

I've had this issue.

Try IT (http://nullinfo.s21.xrea.com/cgi/counter/count.xcg?down=IT_YV12_0103.zip)

[k0r0n4]

Awesome! The included text documents looked like they had some multibyte characters from another language (japanese? chinese?), but I was able to at least copy the parameters out. Works like a charm....I especially like how it automatically decimates :D

Thanks again!

EDIT: Changed my mind, see post titled "One last note..." below this one

[Leak]

Originally posted by k0r0n4
Any ideas on how to fix this, or do I just have to live with the imperfections of automatic IVTC?

Well, there's Tritical's TIVTC (http://forum.doom9.org/showthread.php?s=&threadid=82264&perpage=40&pagenumber=3#post575473) which more or less is a big improvement upon Decomb. It also comes with English documentation... :D

[Nicholi]

If by chance you wanted to continue using Decomb. You can simply just use override files for the few frames you don't like.

Telecide(blahblahblah,ovr="filepath\ovr.txt")

Likewise there is a handy app now, Yatta (http://yatta.mellbin.org), which makes OVR use pretty easy. As well as adding about a dozen other helpful options for IVTC and filtering.

[lamer_de]

In fact, I wouldn't have seen it on that picture if you hadn't pointed it out. How people can spot such things when watching the video at full speed is beyond me...

CU,
lamer_de

[Nicholi]

It's not that hard once you get used to it. And especially easy in animation. Such as her mouth, it is either definitly open, or definitly closed. There is no "in between" frame movement for that animation. The more distant and blurred objects get though, it does become increasingly difficult to judge whether a match is correct or not. Next best thing would be to just find the pattern of the current scene to be sure.

[k0r0n4]

If by chance you wanted to continue using Decomb. You can simply just use override files for the few frames you don't like.
Naw, IT seems to work REALLY well...I just wish I could figure out the author's name so I could give him/her credit.

In fact, I wouldn't have seen it on that picture if you hadn't pointed it out. How people can spot such things when watching the video at full speed is beyond me...
Well, normally I wouldn't have caught it except in that scene there's only one thing moving, and thats her mouth. Yes, I'm one of those people who is obcessive compulsive when it comes to video quality, but even I'm not crazy enough to look at video frame-by-frame, and I don't even slow it down with AssumeFPS to check for bad decombs, it's just I always focus on even the slightest motion when doing IVTC because I *hate* when I see poor quality AMVs when that's the easiest casualty to avoid.

Also, what Nicholi said is true...her mouth only has three positions; open, half-open, and closed. Sometimes when it jumps from closed to open (skipping half-open) or vice-versa, you can see the bad decombing.

In reguard to using overrides, I'd rather not have to bother. IT has yet to produce a single bad comb that I've caught (and, as you can see, I catch very minor flaws), so I might as well not fix it if it's not broken. However, if I do find a few poorly decombed frames, I'll be sure to override them :)

Thanks for the help! :D

EDIT: Changed my mind, see post titled "One last note..." below this one

[mohn]

@ k0r0n4

I know, your problem is already solved, but if you want to use Telecide() and your source material looks very clean,
you should lower the noise threshold to 0 (nt=0).

At least that did the trick for me in many cases.

[k0r0n4]

I know, your problem is already solved, but if you want to use Telecide() and your source material looks very clean, you should lower the noise threshold to 0 (nt=0).

At least that did the trick for me in many cases.

Ahhhh, that solved it. I still get some bad frames near cuts though with telecide, so I'll stick with IT. Thanks :)

EDIT: Changed my mind, see next post

[k0r0n4]

Ok, I just wanted to say one last thing before this thread dies.

I found that in rare cases, IT was producing ghosts in the video. Since I can't read IT's readme, and it didn't seem like there were too many options anyways, I decided to switch back over to the Telecide+Decimate method. Here's the interesting results I found.

I set nt (noise threshold) to 0. I left "Back" equal to 2 (forced backwards matching), but I still got very small errors.

http://img.photobucket.com/albums/v16/yoitsmeremember/Doom9/bad-ivtc2.jpg
Notice the small combed lines on her mouth, as I had before.

I found that by setting back to 0 (off), these bad lines on her mouth would disappear. However, other, more obvious lines would appear.

http://img.photobucket.com/albums/v16/yoitsmeremember/Doom9/bad-ivtc3.jpg

This one is less obvious, but only appears as a result of missing backward matching as well:
http://img.photobucket.com/albums/v16/yoitsmeremember/Doom9/bad-ivtc4.jpg

Now, I tried everything to fix this. Turning the "guide" function on, turning chroma off, turning belding on, etc...and none worked. However, I noticed something when looking at the bad frames. The vmetrics on the frames that were combed with "back" set to 0 were higher than the frames that were combed with "back" set to 2. The solution? Set "back" to 1, and set bthresh to EXACTLY 16 as it turned out.

Now I have a flawless IVTC (as far as I can see) :D

I hope this helps anyone with the same problem in the future.

[tct666]

1.Use copy frame

freezeframe
or
#copy Next frame
function N(clip clip, int frame) {
clip.DeleteFrame(frame).DuplicateFrame(frame)
}
#copy previous frame
function P(clip clip, int frame) {
clip.DeleteFrame(frame).DuplicateFrame(frame-1)
}

2.Use manual IVTC with TmpgEnc

[k0r0n4]

1.Use copy frame

freezeframe
or
#copy Next frame
function N(clip clip, int frame) {
clip.DeleteFrame(frame).DuplicateFrame(frame)
}
#copy previous frame
function P(clip clip, int frame) {
clip.DeleteFrame(frame).DuplicateFrame(frame-1)
}

2.Use manual IVTC with TmpgEnc
I no longer need to copy over bad frames because I'm no longer getting combed frames. Besides, using the "override" feature is more efficient anyways.

Also, I found TMPGEnc was far worse at IVTCing than telicide or IT, no matter what I did to try and tweak it.

[Dreassica]

strange, for me its exactly th eother way round. i spend ages finding settings for telecide and still not getting 100% staisfactory results, while doing manual ivtc in tmpgenc always allows me to do a 100% perfect ivtc.

[tct666]

Originally posted by k0r0n4
I no longer need to copy over bad frames because I'm no longer getting combed frames. Besides, using the "override" feature is more efficient anyways.

Also, I found TMPGEnc was far worse at IVTCing than telicide or IT, no matter what I did to try and tweak it.

The image you posted above is a combed frame.
Try to use manual/auto IVTC with TmpgEnc,and you can find a lot of combed frames which are like the one you posted above,especially the scenes they speak.

[Nicholi]

If you use overrides in Decomb, thats what you might call "manual IVTC". So really, there must not be much of a difference. Unless I am misinterpreting TMPG's option, and it's not really manual.

Using Yatta to generate all your overrides for decomb is extremely helpful and will also be extremely easy to understand if you already know how Telecide/Decimate work in conjunction. Makes it alot simpler then creating overrides yourself with notepad.

[k0r0n4]

Unless I am misinterpreting TMPG's option, and it's not really manual.
Yeah, I was confused as to why people were saying that too.

[tct666]

Originally posted by Nicholi
If you use overrides in Decomb, thats what you might call "manual IVTC". So really, there must not be much of a difference. Unless I am misinterpreting TMPG's option, and it's not really manual.

Using Yatta to generate all your overrides for decomb is extremely helpful and will also be extremely easy to understand if you already know how Telecide/Decimate work in conjunction. Makes it alot simpler then creating overrides yourself with notepad.

30fps->doubleweaved(in TmpgEnc's IVTC window)->60fps->manully/auto IVTC in TmpgEnc and choose 4 frames every 10 frame->24P

Doubleweaving produces all posible frames,and you can do 100% IVTC
with manual IVTC except hybrid and 30i/24P/30P-mixed DVD.

[k0r0n4]

Originally posted by tct666
Doubleweaving produces all posible frames,and you can do 100% IVTC
with manual IVTC except hybrid and 30i/24P/30P-mixed DVD.
Not true. I used AviSynth's DoubleWeave() function, but here's the problem with it.

First of all, this is NOT a hybrid/mixed DVD. I discovered that the pattern was: A1 A2 B1 A2 C1 B2 C1 C2 D1 D2

Now, as you can see, even with DoubleWeave() you cannot get a B1 B2 frame. Obviously, I tried manual IVTC first, but without success. Why? Well, the pattern does not change, but the offset does every time a scene change occurs. Thus in order to do perfect manual IVTC I'd have to go through the entire clip, find where each scene change occurs, and find which offset of the pattern matches that scene.

This is not an option when doing several hours worth of video with reasonably frequent scene changes.

Thus I had to resort to automatic IVTC instead. Sure, it takes quite a while to tweak it to get acceptable results, but it takes FAR less time than going through the video scene-by-scene.

[tct666]

Originally posted by k0r0n4
Not true. I used AviSynth's DoubleWeave() function, but here's the problem with it.

First of all, this is NOT a hybrid/mixed DVD. I discovered that the pattern was: A1 A2 B1 A2 C1 B2 C1 C2 D1 D2

Now, as you can see, even with DoubleWeave() you cannot get a B1 B2 frame. Obviously, I tried manual IVTC first, but without success. Why? Well, the pattern does not change, but the offset does every time a scene change occurs. Thus in order to do perfect manual IVTC I'd have to go through the entire clip, find where each scene change occurs, and find which offset of the pattern matches that scene.

This is not an option when doing several hours worth of video with reasonably frequent scene changes.

Thus I had to resort to automatic IVTC instead. Sure, it takes quite a while to tweak it to get acceptable results, but it takes FAR less time than going through the video scene-by-scene.

Use copy frame or choose the prosssive frame from the (previous/next) scene to copy over the interlaced frame to match Audio-Video in TmpgEnc.

[k0r0n4]

Originally posted by tct666
Use copy frame or choose the prosssive frame from the (previous/next) scene to copy over the interlaced frame to match Audio-Video in TmpgEnc.
:confused: :confused: :confused:

The issue isn't copying progressive frames over interlaced frames or maintaining A/V synch. The issue was that even with DoubleWeave I'd be missing frames, and even when I corrected the field order there would be 5 different offsets due to scene cuts, so I'd still have to go through the entire clip scene-by-scene to get the right offset for each scene.

Which is why to be practical I have to do automatic IVTC.

[tct666]

Originally posted by k0r0n4
:confused: :confused: :confused:

The issue isn't copying progressive frames over interlaced frames or maintaining A/V synch. The issue was that even with DoubleWeave I'd be missing frames, and even when I corrected the field order there would be 5 different offsets due to scene cuts, so I'd still have to go through the entire clip scene-by-scene to get the right offset for each scene.

Which is why to be practical I have to do automatic IVTC.

The pattern is not always rugualr: A1 A2 B1 A2 C1 B2 C1 C2 D1 D2
The DVD's animation may be edited after being telecined.

For example,the scene 1
A1 A2 B1 A2 C1 B2 C1 C2 D1 D2 E1 E2 F1 E2
the scene 2
A1 A2 B1 A2 C1 B2 C1 C2 D1 D2
...

Hence,copying frame must be needed to match A/V.

[k0r0n4]

Originally posted by tct666
The pattern is not always rugualr: A1 A2 B1 A2 C1 B2 C1 C2 D1 D2
The DVD's animation may be edited after being telecined.

For example,the scene 1
A1 A2 B1 A2 C1 B2 C1 C2 D1 D2 E1 E2 F1 E2
the scene 2
A1 A2 B1 A2 C1 B2 C1 C2 D1 D2
...

Hence,copying frame must be needed to match A/V.
No, the pattern is always regular, it's just the offset that is different. As I said, it is not a hybrid clip or a bad telecine, it uses the same pattern and only resets the location of where it is in a pattern due to a scene change, which is what telecined video should do.

Copying frames are not needed whatsoever to maintain A/V synch.

[tct666]

Originally posted by k0r0n4
No, the pattern is always regular, it's just the offset that is different. As I said, it is not a hybrid clip or a bad telecine, it uses the same pattern and only resets the location of where it is in a pattern due to a scene change, which is what telecined video should do.

Copying frames are not needed whatsoever to maintain A/V synch.
If it is regualr,it is so easy to do IVTC.
In the IVTC's window in the TmpgEnc,enter Ctrl-P.
Enter "10100" in the frame named 0a.

0a out frame->1
0b not ->0
1a out frame->1
1b not ->0
2a not ->0

2b out frame->1
3a not ->0
3b out frame->1
4a not ->0
4b not ->0
.
.
.

[Dreassica]

And that will only work until the first scenechange, then he will have to shift the pattern again, in his case at every scenechange, so u can understand that's allot of work to do. If he isn't willing to do that, he's better of using an auto ivtc config using telecide/decimate since its much better at auto then tmpgenc's.

[k0r0n4]

Originally posted by Dreassica
And that will only work until the first scenechange, then he will have o shift the pattern again, in his case at every scenechange, so u can understand tahst allot of work to do. If he isn't illing to do that, hes better of using an auto ivtc config using telecide/decimate since its much better at auto then tmpgencs.
exactly :D :D :D

[Chainmax]

I have to second the recommendation for TIVTC. Use mode=2 in TFM and you can kiss mouth combs bye-bye :).

[k0r0n4]

Originally posted by Chainmax
I have to second the recommendation for TIVTC. Use mode=2 in TFM and you can kiss mouth combs bye-bye :).
TFM?

[ObiKenobi]

Originally posted by k0r0n4
TFM?

That's the frame matching function within TIVTC.

[k0r0n4]

Ahh, thanks! TIVTC works even better than Telicide and requires no tweaking! Awesome :D