What is this issue?

[Akai-Shuichi]

https://forum.videohelp.com/images/imgfiles/NZeldD4.png

around the outer edge of her left arm and between her breasts.

I can't tell if this is aliasing or something not picked up by IVTC.

[poisondeathray]

looks like residual combing - e.g when fields aren't perfectly aligned

you can adjust the field matcher settings (maybe combing thresholds), or apply something afterwards to deal with it like vinverse

[jagabo]

The lines are thicker and darker in one field than the other. A simple bob:



I suspect the IVTC took fields from two different video frames and the encoding or recording of the source thinned or thickened lines in on of the fields. Or maybe the fields really come from different film frames and the IVTC picked the wrong fields. You'd have to include more frames to tell which.

[Akai-Shuichi]

The lines are thicker and darker in one field than the other. A simple bob:


[Attachment 30548 - Click to enlarge]

I suspect the IVTC took fields from two different video frames and the encoding or recording of the source thinned or thickened lines in on of the fields. Or maybe the fields really come from different film frames and the IVTC picked the wrong fields. You'd have to include more frames to tell which.

It looks like that the lines are thicker/thinner, you are correct. What would you do in this situation?

using daa() seems to fix it, but I don't know if its a good idea to use that on the whole source when its just certain frames that are like that.

I used TFM().TDecimate() in the first place. If there is another way to IVTC this that I could try, I'm all ears.

[themaster1]

TFM(order=1) # order=1 (TFF)
vinverse # to remove residual combing
TDecimate(mode=1) #for anime

[Akai-Shuichi]

vinverse doesn't seem to be doing anything to it.

[poisondeathray]

You can adjust the settings.

vinverse2(sstr=0.5)

You said DAA worked - try out some other AA filters as well

If you want to apply some filter to only to those frames, and only to specific areas within those frames, you'd have to create a selection mask . Very difficult to do accurately on something like this

Or maybe there are other fields that are a better match, and you can adjust the TFM settings for better results

[Akai-Shuichi]

QTGMC(Preset="medium", EZDenoise=1.0)
SelectEven()
TDecimate(mode=1)

seems to get the job done perfectly.

[jagabo]

Try TFM(mode=0) and see if you get less of them.

QTGMC() is a very slow solution. It's also likely to create other artifacts occasionally.

[Akai-Shuichi]

Try TFM(mode=0) and see if you get less of them.

QTGMC() is a very slow solution. It's also likely to create other artifacts occasionally.

tfm(mode=0) didn't seem to work. Would daa() be a better solution than QTGMC() then?

Here is a clip of it, see if you have any better luck: https://mega.co.nz/#!tUpFjbQC!msquZDZyXLwU2ZyY06GGtI5jlBsffurOvYLOaki_nAo

[jagabo]

poisondeathray's vinverse2(sstr=0.5) works pretty well on the sample image. daa() works well and darkens the line a bit.

[Akai-Shuichi]

poisondeathray's vinverse2(sstr=0.5) works pretty well on the sample image. daa() works well and darkens the line a bit.

Ah, wow it works! Thanks. I'll stick with vinverse for taking of it since its probably safer.

[poisondeathray]

But I think it leaves some aliasing on some frames near the end on the clip. A generic AA function like daa3() , santiag() etc... will look better on some frames, but perhaps worse on others


Or you can apply it conditionally, only on frames that have combing detected, by adjusting TFM parameters, like cthresh, MI, clip2 etc... so only those frames that are detected as combed will be swapped with "clip2" - thus limiting the "damage" even more . This technique can be used for any filter or processing - "clip2" can be QTGMC for example , or anything to "replace" wherever combing is detected. If you wanted to use daa3() or some other AA you could do it that way

eg.

MPEG2Source("vtsclip.d2v", cpu=0)
orig=last

orig
AssumeTFF()
TFM()
vinverse2(sstr=0.5)
vin=last

orig
TFM(mode=0, PP=2, cthresh=2, mi=40, mChroma=false, blockx=16, blocky=16, display=true, clip2=vin)
Tdecimate()

That example uses vinverse2 for clip2, so it will indicate whenever a field match has combing detected because display=true it will be written in an overlay, and replace that frame with one that has vinverse2 applied. The other frames won't get vinverse2 applied

set display=false when you want to actually use it, and use pp=0 to help debug it (turns post processing off). So tweak the thresholds until you are satisfied that combing is detected properly on other sections as well

[Akai-Shuichi]

But I think it leaves some aliasing on some frames near the end on the clip. A generic AA function like daa3() , santiag() etc... will look better on some frames, but perhaps worse on others


Or you can apply it conditionally, only on frames that have combing detected, by adjusting TFM parameters, like cthresh, MI, clip2 etc... so only those frames that are detected as combed will be swapped with "clip2" - thus limiting the "damage" even more . This technique can be used for any filter or processing - "clip2" can be QTGMC for example , or anything to "replace" wherever combing is detected. If you wanted to use daa3() or some other AA you could do it that way

eg.

MPEG2Source("vtsclip.d2v", cpu=0)
orig=last

orig
AssumeTFF()
TFM()
vinverse2(sstr=0.5)
vin=last

orig
TFM(mode=0, PP=2, cthresh=2, mi=40, mChroma=false, blockx=16, blocky=16, display=true, clip2=vin)
Tdecimate()

That example uses vinverse2 for clip2, so it will indicate whenever a field match has combing detected because display=true it will be written in an overlay, and replace that frame with one that has vinverse2 applied. The other frames won't get vinverse2 applied

set display=false when you want to actually use it, and use pp=0 to help debug it (turns post processing off). So tweak the thresholds until you are satisfied that combing is detected properly on other sections as well

I can't seem to get it to use vinverse...it detects them, as pp=2 and pp=6 deals with them, but when i set it to pp=0 it doesn't use vinverse

[poisondeathray]

Yes, that's the point - you only use pp=0 to debug the script. You adjust it back when you actually use it. Since there is no post processing when pp=0, you should see the combing. You have to be able to see the combing to adjust the values and customize the script. Then you adjust the thresholds until frames that have combing are detected, and those that don't have combing are not detected. You don't want to process frames that are clean because you will degrade them unnecessarily - that's the whole idea

The overlay will say something like "combing detected! deinterlaced" or something like that. If you see combing, but it doesn't say "detected", then you need to adjust the threshold. Similarly, if it says combing detected, when there is none, that's a false positive, and you have to adjust the values.

When you use clip2, it applies whatever process is specified by clip2, whether it's deinterlacing , AA etc... whatever you want. It just replaces those frames that are detected as "combed" with clip2

[Akai-Shuichi]

Yes, that's the point - you only use pp=0 to debug the script. You adjust it back when you actually use it. Since there is no post processing when pp=0, you should see the combing. You have to be able to see the combing to adjust the values and customize the script. Then you adjust the thresholds until frames that have combing are detected, and those that don't have combing are not detected. You don't want to process frames that are clean because you will degrade them unnecessarily - that's the whole idea

The overlay will say something like "combing detected! deinterlaced" or something like that. If you see combing, but it doesn't say "detected", then you need to adjust the threshold. Similarly, if it says combing detected, when there is none, that's a false positive, and you have to adjust the values.

When you use clip2, it applies whatever process is specified by clip2, whether it's deinterlacing , AA etc... whatever you want. It just replaces those frames that are detected as "combed" with clip2

This is what I put in:

MPEG2Source("j.d2v", cpu2="ooooxx")
vin= vinverse(sstr=0.5)
TFM(mode=0, cthresh=2, mi=40, pp=1, mChroma=false, display=true, blockx=16, blocky=16, clip2=vin)
TDecimate(mode=1)

It is picking them up perfectly in the whole video, but doesn't actually seem to be using the vinverse()

One more thing, in your script what is the point of having TFM() 2 times?

[poisondeathray]

This is what I put in:

MPEG2Source("j.d2v", cpu2="ooooxx")
vin= vinverse(sstr=0.5)
TFM(mode=0, cthresh=2, mi=40, pp=1, mChroma=false, display=true, blockx=16, blocky=16, clip2=vin)
TDecimate(mode=1)

It is picking them up perfectly in the whole video, but doesn't actually seem to be using the vinverse()

One more thing, in your script what is the point of having TFM() 2 times?

You currently have pp=1, so that means "don't deinterlace", and/or don't use clip2

Even if you use pp=2, the way you currently have it, vinverse is applied to the video before TFM, so it's applied to the video while it's still interlaced, not field matched or progressive - big problems

You can think of it as 2 versions of the video. One is normal TFM.TDecimate(). The other one has vinverse applied to ALL the frames. It swaps between the two only on frames where combing is detected. So uncombed frames are passed through, and only combed frames get vinverse treatment (or whatever treatment you specify for clip2)

[Akai-Shuichi]

This is what I put in:

MPEG2Source("j.d2v", cpu2="ooooxx")
vin= vinverse(sstr=0.5)
TFM(mode=0, cthresh=2, mi=40, pp=1, mChroma=false, display=true, blockx=16, blocky=16, clip2=vin)
TDecimate(mode=1)

It is picking them up perfectly in the whole video, but doesn't actually seem to be using the vinverse()

One more thing, in your script what is the point of having TFM() 2 times?

You currently have pp=1, so that means "don't deinterlace", and/or don't use clip2

Even if you use pp=2, the way you currently have it, vinverse is applied to the video before TFM, so it's applied to the video while it's still interlaced, not field matched or progressive - big problems

You can think of it as 2 versions of the video. One is normal TFM.TDecimate(). The other one has vinverse applied to ALL the frames. It swaps between the two only on frames where combing is detected. So uncombed frames are passed through, and only combed frames get vinverse treatment (or whatever treatment you specify for clip2)

Alright, I get it now. for the orig part is necessary to keep using orig or would it be alright to just leave the initial MPEG2Source in the beginning?

[poisondeathray]

Alright, I get it now. for the orig part is necessary to keep using orig or would it be alright to just leave the initial MPEG2Source in the beginning?

No you don't need to if you have the correct implied "last". You can organize scripts however you want to

I explicitly split out sections and call out named variables to keep everything clean and organized and easier to understand what is going on. Often with more complex scripts than this, you can get confused or lose track of what implied "last" was and get the wrong results.

[Akai-Shuichi]

Thanks a lot man. Helped me learn a lot too!