2 questions about IVTC in Avisynth

[Megafox]

1. Why does not TIVTC notice a small comb with a height less than 10 pixels? I configured everything as accurately as possible, good frames are not affected anymore, but some interlaced remain:
TFM (slow = 2, cthresh = 1, scthresh = 1, MI = 1000, PP = 0, mode = 7, mChroma = false, mmsco = false , metric = 1)
TDecimate (hybrid = 1).
2. Which plug-in will search for interlaced frames not by detecting the comb, but by comparing the fields?

[johnmeyer]

1. I don't know why you started another thread. The one you posted earlier today is the same subject.

2. Since you haven't posted a sample, I can't help you and I suspect others will have the same problem.

3. What test are you doing to determine whether it is interlaced?

4. Finally, technically speaking, telecined video is not interlaced because fields are duplicated and therefore are from the same moment in time as the original fields from which they were copied. By contrast, true interlaced video has fields that are always from different moments in time. As a result, IVTC is NOT the same thing as deinterlacing. If your video is truly interlaced, then IVTC is the wrong tool to use.

Which brings me back to #2: post a sample.

[poisondeathray]

1. Why does not TIVTC notice a small comb with a height less than 10 pixels? I configured everything as accurately as possible, good frames are not affected anymore, but some interlaced remain:
TFM (slow = 2, cthresh = 1, scthresh = 1, MI = 1000, PP = 0, mode = 7, mChroma = false, mmsco = false , metric = 1)
TDecimate (hybrid = 1).

Did you look at the READ ME file ?

blockx -

Sets the x-axis size of the window used during combed frame detection. This has
to do with the size of the area in which MI number of pixels are required to be
detected as combed for a frame to be declared combed. See the MI parameter
description for more info. Possible values are any number that is a power of
2 starting at 4 and going to 2048 (i.e. 4, 8, 16, 32, ... 2048).

Default: 16 (int)


blocky -

Sets the y-axis size of the window used during combed frame detection. This has
to do with the size of the area in which MI number of pixels are required to be
detected as combed for a frame to be declared combed. See the MI parameter
description for more info. Possible values are any number that is a power of
2 starting at 4 and going to 2048 (i.e. 4, 8, 16, 32, ... 2048).

Default: 16 (int)

blockx =16, blocky = 16 by default . That is a 16x16 square. You would need blocky <10

But you have PP=0 , so that means even if it detects combing, there is no post processing . Instead of using the dumb cubic post processing, many people choose to specify a clip2 parameter from which it takes higher quality deinterlaced frames from. It applies only if combing is detected



If you can't figure it out, or something else is going on, or can't describe it properly, post a sample

[manono]

TFM (slow = 2, cthresh = 1, scthresh = 1, MI = 1000, PP = 0, mode = 7, mChroma = false, mmsco = false , metric = 1)
TDecimate (hybrid = 1).

As pdr pointed out, your script has TFM do no post-processing at all. In addition, your CThresh=1 forces TFM to find almost all frames as interlaced and to post-process them. That combined with the PP=0 should have clued you in that your script is very peculiar.

2. Which plug-in will search for interlaced frames not by detecting the comb, but by comparing the fields?

Just teach yourself to use TIVTC properly.

And, as others have mentioned, questions such as yours should always be accompanied by a sample.

[hello_hello]

1. Why does not TIVTC notice a small comb with a height less than 10 pixels? I configured everything as accurately as possible, good frames are not affected anymore, but some interlaced remain:
TFM (slow = 2, cthresh = 1, scthresh = 1, MI = 1000, PP = 0, mode = 7, mChroma = false, mmsco = false , metric = 1)
TDecimate (hybrid = 1).

Why don't you upload a sample? If you did, someone might be able to give you the settings to use. It might even allow someone to answer the correct questions, even if you're asking the wrong ones.

2. Which plug-in will search for interlaced frames not by detecting the comb, but by comparing the fields?

TFM is a field matcher, so it does lots of field comparing. After it's matched fields to create whole frames, it then checks for combing. How would it determine there's combing without checking for combing?
In a perfect world you'd be able to match the telecined fields and there'd never be any combing because you're simply putting progressive frames back together again, but it's not a perfect world. For TFM, you can increase the combing detection sensitivity until every pixel is considered combed, if that's what it takes, but usually there's no need to go that far.

My guess is you're not seeing combing, and chances are it's dot crawl. If that's the case, all the de-interlacing in the world won't fix it. If you upload a sample someone can tell you for sure. If not, you might get lucky and one of the regulars will be happy to play "guess the filter" with you, but without a sample it's probably a waste of time.

By the way, an easy way to see what's being detected as combed is to do something like this:

A=BlankClip(Last, Color=color_pink).KillAudio()
TFM(Clip2=A)
TDecimate()

Instead of de-interlacing, TFM will replace any combing it detects with parts of clip A's frames, but in the above example clip "A" contains nothing but whole frames of pink. So where you see pink in the video, that's where TFM detects combing.
Actually.... if PP= 2, 3, or 4, the whole frame will be detected as combed if any combing is detected, for pp=5, 6, or 7, it'll be just the combed sections, and for pp= 0 or 1, it won't do anything.

The screenshot below used the default TFM settings. It detects more combing than my eyes unless I upscale the video quite a bit, which is one reason I suspect your problem mightn't be combing. Please upload a sample.

[Megafox]

1. Why does not TIVTC notice a small comb with a height less than 10 pixels? I configured everything as accurately as possible, good frames are not affected anymore, but some interlaced remain:
TFM (slow = 2, cthresh = 1, scthresh = 1, MI = 1000, PP = 0, mode = 7, mChroma = false, mmsco = false , metric = 1)
TDecimate (hybrid = 1).

Why don't you upload a sample? If you did, someone might be able to give you the settings to use. It might even allow someone to answer the correct questions, even if you're asking the wrong ones.

2. Which plug-in will search for interlaced frames not by detecting the comb, but by comparing the fields?

TFM is a field matcher, so it does lots of field comparing. After it's matched fields to create whole frames, it then checks for combing. How would it determine there's combing without checking for combing?
In a perfect world you'd be able to match the telecined fields and there'd never be any combing because you're simply putting progressive frames back together again, but it's not a perfect world. For TFM, you can increase the combing detection sensitivity until every pixel is considered combed, if that's what it takes, but usually there's no need to go that far.

My guess is you're not seeing combing, and chances are it's dot crawl. If that's the case, all the de-interlacing in the world won't fix it. If you upload a sample someone can tell you for sure. If not, you might get lucky and one of the regulars will be happy to play "guess the filter" with you, but without a sample it's probably a waste of time.

By the way, an easy way to see what's being detected as combed is to do something like this:

A=BlankClip(Last, Color=color_pink).KillAudio()
TFM(Clip2=A)
TDecimate()

Instead of de-interlacing, TFM will replace any combing it detects with parts of clip A's frames, but in the above example clip "A" contains nothing but whole frames of pink. So where you see pink in the video, that's where TFM detects combing.
Actually.... if PP= 2, 3, or 4, the whole frame will be detected as combed if any combing is detected, for pp=5, 6, or 7, it'll be just the combed sections, and for pp= 0 or 1, it won't do anything.

The screenshot below used the default TFM settings. It detects more combing than my eyes unless I upscale the video quite a bit, which is one reason I suspect your problem mightn't be combing. Please upload a sample.

Why are not there such deinterlacers? I have already invented an algorithm, but I can not find such a deinterlacer, I'll try to compile it myself and check how it works. Here's what it does:
# an interlaced frame that one between progressive frames has a bottom field from the previous frame and an upper field from the next frame
# If there are two interlaced frames between the progressive frames, first interlaced frame have bottom field will correspond to the bottom field of the previous progressive frame , and the second frame have top field, like the next progressive frame
# If 3 or more interlaced frames, the first interlaced frame will have the lower field from the previous progressive frame, and the last have the top field from the next progressive frame, the interlaced frames inside are considered interlaced.
First, by comparing the fields (by pixels), the first point (an interlaced frame that has a bottom field identical to the lower field of the previous one) is found, after this frame is found, the second point is searched, re-checking the current frame and the following (a frame whose top field is completely identical to the top field of the next frame), then:
1) If both points are marked in one frame, it is considered interlaced, and matched fields for it, and there is a switch to one frame ahead
2) If the first and second point in two adjacent frames, then there is a matching for these two frames and switching the search to the next after them
3) If two points are in frames with yet another third interlaced Inside, these 3 frames also receive their fields, and the search engine checks the following frames. If you suddenly get absolutely identical frames, nothing will change if the first one is considered interlaced.
Translated by Google

[hello_hello]

There's probably something being lost in the translation, but what you've described sounds like field matching. That's what TFM does. It's what the "FM" in TFM stands for. Field Matching. De-interlacing and field matching are two completely different things.

At it's simplest, a field matcher matches the bottom field of each frame with the top field of the same frame and the top field of the next frame. The best match (least amount of combing) becomes a new progressive frame. If a frame is truly unique and progressive, only it's own fields can match each other. If there's combing because of telecine and it's fields don't match, the bottom field should match the top field of the next frame, and so on and so on. When there's no good matches for a cycle or two, the source is no longer considered "film" and it's de-interlaced as "video" (for hybrid sources). When matches can be found again, it's again treated as film.

TFM can be more clever, and depending on the mode it'll try to match fields from the previous frame as well as the following frame. In some circumstances it can even try to match bottom fields with bottom fields etc. In other words, it does it's best to output clean progressive frames, but when there's no matches without combing, which can happen for various reasons, it has to resort to de-interlacing to fix it.

Anyway, given you've ignored several requests for a sample and you seem to have decided what the problem is and how to fix it, this looks like it'll be one of those threads that'll keep going round in circles without achieving anything, so I'll leave you to it.

[Megafox]

There's probably something being lost in the translation, but what you've described sounds like field matching. That's what TFM does. It's what the "FM" in TFM stands for. Field Matching. De-interlacing and field matching are two completely different things.

At it's simplest, a field matcher matches the bottom field of each frame with the top field of the same frame and the top field of the next frame. The best match (least amount of combing) becomes a new progressive frame. If a frame is truly unique and progressive, only it's own fields can match each other. If there's combing because of telecine and it's fields don't match, the bottom field should match the top field of the next frame, and so on and so on. When there's no good matches for a cycle or two, the source is no longer considered "film" and it's de-interlaced as "video" (for hybrid sources). When matches can be found again, it's again treated as film.

TFM can be more clever, and depending on the mode it'll try to match fields from the previous frame as well as the following frame. In some circumstances it can even try to match bottom fields with bottom fields etc. In other words, it does it's best to output clean progressive frames, but when there's no matches without combing, which can happen for various reasons, it has to resort to de-interlacing to fix it.

Anyway, given you've ignored several requests for a sample and you seem to have decided what the problem is and how to fix it, this looks like it'll be one of those threads that'll keep going round in circles without achieving anything, so I'll leave you to it.

This is not a matchion, but a comparison of fields, and TFM uses non-exact methods, my method should see all the interlaced frames, but I'm unfortunately not the developer to check it

[jagabo]

Megafox, your analysis is very naive. Tritical is put more thought and effort into this than you could ever hope to. Real video is full of complications that make this far more difficult than you think.

Your source video may have orphaned fields -- fields for which there is no complimentary field to reproduce the original film frame. This is very common with video that was telecined then edited, sped up, or slowed down. Or you may have field blending from a frame rate conversion. Since you refuse to provide a sample nobody is going to be able to help you.

[Megafox]

Megafox, your analysis is very naive. Tritical is put more thought and effort into this than you could ever hope to. Real video is full of complications that make this far more difficult than you think.

Your source video may have orphaned fields -- fields for which there is no complimentary field to reproduce the original film frame. This is very common with video that was telecined then edited, sped up, or slowed down. Or you may have field blending from a frame rate conversion. Since you refuse to provide a sample nobody is going to be able to help you.

And I'm not talking about such videos that have lost the fields, I'm talking about telecined video, and in the video after the transformations even TFM did not help me. There is nothing to restore

[johnmeyer]

Anyway, given you've ignored several requests for a sample and you seem to have decided what the problem is and how to fix it, this looks like it'll be one of those threads that'll keep going round in circles without achieving anything, so I'll leave you to it.

Unfortunately, I sense that you are correct.

The bad thing about posts like this is that it makes me much less likely to help the next person. There are just too many threads like this that are exactly like you described this one.

Perhaps the sysops could enforce a rule that if more than one person requests a clip or requests that the person post their script, and they don't, the thread gets locked. Do it more than three times, you can't post again for a few months. I am tired of wasting my time with people like this.

[jagabo]

The OP doesn't understand that the possibilities are so numerous that nobody is going to write a post describing all the possible fixes. Uploading a sample video would allow someone here to narrow down problem and provide a fix, or an explanation of why there is no perfect fix. No doubt the video is hentai so he's embarrassed to post a sample.

[hello_hello]

The bad thing about posts like this is that it makes me much less likely to help the next person. There are just too many threads like this that are exactly like you described this one.

I wonder if Katie (doom9) has a brother or sister? This thread is much like many of hers, only with a slightly less abusive attitude when the answers given aren't the answers she decided in advance they should be.

[johnmeyer]

Katie is unique (thankfully).

In her defense (I can't believe I just wrote that) she also has occasionally demonstrated some competence. Her main problem is her abusiveness. She reminds me of a cat my sister once had where you would pet it, and it would start to purr, but if you petted it for more than twenty seconds, it would turn and bite you. Lovely.