TDeint ELA1 configuration

[Megafox]

In some episodes of my series, I found groups of frames with lost fields, only in places without movement. I was looking for a solution for a long time, until I found ELA. It works better than any photo enhancement program, but I can not get rid of some defects. After frame is restored from field, on some lines appear dots, which should not be there. How can I fix this and make priority of the lines higher? Example: if shade of line is near to main color than dots, then ignore them.

Script: TDeint(mode=-1)

[poisondeathray]

In some episodes of my series, I found groups of frames with lost fields, only in places without movement. I was looking for a solution for a long time, until I found ELA. It works better than any photo enhancement program, but I can not get rid of some defects. After frame is restored from field, on some lines appear dots, which should not be there. How can I fix this and make priority of the lines higher? Example: if shade of line is near to main color than dots, then ignore them.

Script: TDeint(mode=-1)

You can probably discard it in your case. Because in scenes with no movement, then by definition all those frames should have been identical. Thus you can replace with a neighboring good frame

[jagabo]

You should probably be using an inverse telecine anyway. Upload a sample of your source video.

[Megafox]

In some episodes of my series, I found groups of frames with lost fields, only in places without movement. I was looking for a solution for a long time, until I found ELA. It works better than any photo enhancement program, but I can not get rid of some defects. After frame is restored from field, on some lines appear dots, which should not be there. How can I fix this and make priority of the lines higher? Example: if shade of line is near to main color than dots, then ignore them.

Script: TDeint(mode=-1)

You can probably discard it in your case. Because in scenes with no movement, then by definition all those frames should have been identical. Thus you can replace with a neighboring good frame

You did not understand me. I explain. I made comparsion to see, how deinterlacer copes with the increase of resolution. This frame is not orphaned. I attached a video, where fragment clearly shows how resolution falls in all places of frame except mouth, even noise in those places does not move. Some bug of telecine. Only way to fix this is ELA. So I ask how to configure TDeint(mode=-1) so that there are no interrupted lines (see in comparsion). I hope you understand.

[Megafox]

You should probably be using an inverse telecine anyway. Upload a sample of your source video.

You also did not understand me, read my reply

[poisondeathray]

[
You did not understand me. I explain. I made comparsion to see, how deinterlacer copes with the increase of resolution. This frame is not orphaned. I attached a video, where fragment clearly shows how resolution falls in all places of frame except mouth, even noise in those places does not move. Some bug of telecine. Only way to fix this is ELA. So I ask how to configure TDeint(mode=-1) so that there are no interrupted lines (see in comparsion). I hope you understand.

Not a bug of telecine if that is representative of your source. That's a problem with the source


Why not use some antialiasing combination instead on affected frames ? You'll will get better results than tdeint(mode=-1) ELA (ELA is not the "only way to fix this")

Post your actual source; or what script are you currently using for that AVI ?

Or maybe I'm still misunderstanding what you're trying to say?

[Megafox]

[
You did not understand me. I explain. I made comparsion to see, how deinterlacer copes with the increase of resolution. This frame is not orphaned. I attached a video, where fragment clearly shows how resolution falls in all places of frame except mouth, even noise in those places does not move. Some bug of telecine. Only way to fix this is ELA. So I ask how to configure TDeint(mode=-1) so that there are no interrupted lines (see in comparsion). I hope you understand.

Not a bug of telecine if that is representative of your source. That's a problem with the source


Why not use some antialiasing combination instead on affected frames ? You'll will get better results than tdeint(mode=-1) ELA (ELA is not the "only way to fix this")

Post your actual source; or what script are you currently using for that AVI ?

Or maybe I'm still misunderstanding what you're trying to say?

Problem with source? It was in original video (VOB), script only load d2v and Trim. I cant upload full video (copyright and big size). No, you understand me, I want to increase resolution. What is antialiasing? How to apply it?

[jagabo]

Try:

AviSource("video.avi", pixel_type="YV12")
AssumeTFF()
QTGMC(preset="fast") # or as slow as you can stand
SelectEven()
TDecimate()

To make an un-reencoded clip load one of your VOB files into DgIndex, mark a short section, then File -> Save Project And Demux Video. Upload the m2v.

[Megafox]

Try:

AviSource("video.avi", pixel_type="YV12")
AssumeTFF()
QTGMC(preset="fast") # or as slow as you can stand
SelectEven()
TDecimate()

To make an un-reencoded clip load one of your VOB files into DgIndex, mark a short section, then File -> Save Project And Demux Video. Upload the m2v.

I uploaded clip! Watch top reply

[poisondeathray]

1) If you wanted to post a sample, use dgindex to cut a sample. It would be much smaller than the one you uploaded, and the actual original bitstream.

2) It's not "no movement" . Your description is incorrect. It would be more correct to say the there is no movement in the background and parts of the foreground, but Optimus' mouth is moving in that close up section.

3) The 1st part of screenshot is incorrect. "Original field" should be 720x240 . You have a resized field. Depending on what resizer was used, you will get different results.

4) Read up on "antialiasing". Basically, in this context - it's addressing the "jaggies" or "stair stepping". You mentioned "uninterrupted lines" - that's what you're trying to do with antialiasing - repair the jaggy lines. Notice how Optimus has dotted or jaggy lines in that close up middle section ? The lines are not clean and straight. The problem is in your source. It occurred before or during the DVD transfer.

Avisynth alone has many different AA filters , many different settings, some are geared to different types of aliasing. There are other types of AA algorithms in other programs too .

You can apply them just like any other avisynth filter. AA filters can be very damaging, so you would selectively apply them to affected frames . You might apply different AA filters or settings to different ranges frames. (Some filters are are for finer aliasing , thinner lines; some are more effective on thicker , coarse lines . Sometimes you might want to thin the lines temporarily, apply the filters, then thicken and darken them later)


e.g you want to IVTC most of it without post processing, but apply some AA only to the close up section with optimus talking

o=ffvideoSource("video.avi")

o
assumetff()
tfm(pp=0).tdecimate()
i=last 

a1=i.daa3().maa2()

i.trim(0,62) ++\
a1.trim(63,118) ++\
i.trim(119,0)

You can have different types of AA, different settings etc... Some frames might need a weaker AA, some stronger, some different settings, sometimes something slightly different like residual combing repair . For example ivtc'ed frame #23 has some combing. You might apply something different to fix that frame

[Megafox]

1) If you wanted to post a sample, use dgindex to cut a sample. It would be much smaller than the one you uploaded, and the actual original bitstream.

2) It's not "no movement" . Your description is incorrect. It would be more correct to say the there is no movement in the background and parts of the foreground, but Optimus' mouth is moving in that close up section.

3) The 1st part of screenshot is incorrect. "Original field" should be 720x240 . You have a resized field. Depending on what resizer was used, you will get different results.

4) Read up on "antialiasing". Basically, in this context - it's addressing the "jaggies" or "stair stepping". You mentioned "uninterrupted lines" - that's what you're trying to do with antialiasing - repair the jaggy lines. Notice how Optimus has dotted or jaggy lines in that close up middle section ? The lines are not clean and straight. The problem is in your source. It occurred before or during the DVD transfer.

Avisynth alone has many different AA filters , many different settings, some are geared to different types of aliasing. There are other types of AA algorithms in other programs too .

You can apply them just like any other avisynth filter. AA filters can be very damaging, so you would selectively apply them to affected frames . You might apply different AA filters or settings to different ranges frames. (Some filters are are for finer aliasing , thinner lines; some are more effective on thicker , coarse lines . Sometimes you might want to thin the lines temporarily, apply the filters, then thicken and darken them later)


e.g you want to IVTC most of it without post processing, but apply some AA only to the close up section with optimus talking

o=ffvideoSource("video.avi")

o
assumetff()
tfm(pp=0).tdecimate()
i=last 

a1=i.daa3().maa2()

i.trim(0,62) ++\
a1.trim(63,118) ++\
i.trim(119,0)

You can have different types of AA, different settings etc... Some frames might need a weaker AA, some stronger, some different settings, sometimes something slightly different like residual combing repair . For example ivtc'ed frame #23 has some combing. You might apply something different to fix that frame

In previous comparison, field was duplicated, rather than stretched. I did this to make it easier to compare, and image was not flattened. Since I do not speak English well, I use an translater, and not all the phrases can correctly convey.

I compared many different anti-aliasing filters, different combinations, but combination maa2+daa3 was the best. But she does not see dotted and round lines. I made another comparison, which shows how TDeint detects dotted lines and blurs them, but maa2+daa3 does nothing. TDeint was worse than the rest in quality, but he best knows, how to detect lines.

How can I teach maa2+daa3 to correct dotted, round lines and remaining dentition?

In comparsion work with frame #100.

TDeint command: Trim(100, 100).SeparateFields().Trim(0, 0).TDeint(mode=-1).AssumeFPS(24)

maa2daa3 command: Trim(100, 100).daa3().maa2().AssumeFPS(24)

[poisondeathray]

Post your FULL script. Are you IVTC ing first ?

TDeint on separated fields will give you problems (up/down). Look in motion too , not only still frames

There is a difference in order too - maa2().daa3() is different than daa3().maa2()

If it's not strong enough, you can adjust the settings, or change the filters (there are dozens of AA filters in avisynth, and other programs have additional ones, GPU accelerated). I mentioned one approach - You can thin the lines (awarpsharp2) then AA filters will pick them up more. But AA filters are very very damaging, you have to be careful - they can easily cause more damage than repair . If you wanted to mix/match you can combine it with a mask, but it will probably need lots of tweaking if you use avisynth alone

If you wanted it 100% perfect, I use AE or photoshop. No avisynth filter will give you as good results as that. Because of the completely still BG elements in that shot, you can composite/stretch the BG repair over 55 frames instead of having to do each frame. You might still have to adjust a bit of the mouth a bit

[Megafox]

Post your FULL script. Are you IVTC ing first ?

TDeint on separated fields will give you problems (up/down). Look in motion too , not only still frames

There is a difference in order too - maa2().daa3() is different than daa3().maa2()

If it's not strong enough, you can adjust the settings, or change the filters (there are dozens of AA filters in avisynth, and other programs have additional ones, GPU accelerated). I mentioned one approach - You can thin the lines (awarpsharp2) then AA filters will pick them up more. But AA filters are very very damaging, you have to be careful - they can cause more damage than repair . If you wanted it 100% perfect, I use AE or photoshop - 100% perfect repair everytime . No avisynth filter will give you as good results as that. It's actually easier here, because of the completely still BG elements (you can composite/stretch the BG repair over 55 frames)

FFVideoSource("video.avi")
c0 = Trim(100, 100).AssumeFPS(24)
c1 = c0.daa3().maa2().AssumeFPS(24)
c2 = TDeint(c0.SeparateFields().Trim(0, 0), mode=-1).AssumeFPS(24)
return c0.Subtitle("TDeint") + c1.Subtitle("daa3+maa2")

[poisondeathray]

FFVideoSource("video.avi")
c0 = Trim(100, 100).AssumeFPS(24)
c1 = c0.daa3().maa2().AssumeFPS(24)
c2 = TDeint(c0.SeparateFields().Trim(0, 0), mode=-1).AssumeFPS(24)
return c0.Subtitle("TDeint") + c1.Subtitle("daa3+maa2")

You will go out of sync . The length will be wrong. You'd need to IVTC first

You won't be able to use TDeint on separated fields like that on the actual video - you will have an up/down motion (even/odd fields are spatially offset)


Check it like you are using the full "video.avi" . Don't trim it down

post your FULL script, like you would be using it for the video

[Megafox]

FFVideoSource("video.avi")
c0 = Trim(100, 100).AssumeFPS(24)
c1 = c0.daa3().maa2().AssumeFPS(24)
c2 = TDeint(c0.SeparateFields().Trim(0, 0), mode=-1).AssumeFPS(24)
return c0.Subtitle("TDeint") + c1.Subtitle("daa3+maa2")

You will go out of sync . The length will be wrong. You'd need to IVTC first

You won't be able to use TDeint on separated fields like that on the actual video - you will have an up/down motion (even/odd fields are spatially offset)


Check it like you are using the full "video.avi" . Don't trim it down

post your FULL script, like you would be using it for the video

Why do you need a full script? There's nothing interesting. I do not need to talk about IVTC and synchronization, I know about it without you. Now I cut out one frame FOR TEST, until I can get rid of dotted lines. Give me an example of setting that fixes dotted-lines, I already tried all the settings and can't fix this. Your aWarpSharp2 does nothing except thinning lines. Where did you get that it connects the dotted lines?

[poisondeathray]

Why do you need a full script? There's nothing interesting. I do not need to talk about IVTC and synchronization, I know about it without you. Now I cut out one frame FOR TEST, until I can get rid of dotted lines. Give me an example of setting that fixes dotted-lines, I already tried all the settings and can't fix this. Your aWarpSharp2 does nothing except thinning lines. Where did you get that it connects the dotted lines?

When you use tdeint on separated fields the way you have it - you will get up/down - it's unusable because of the offset. I guess you'll find out the hard way...
I said use awarpsharp2 to thin lines, so you can use some other filters (too thick/too coarse will not be detected by the filter)
You can use luma masks to mix/match areas to fix

But if you want it fixed 100% perfectly, use after effects or photoshop. You won't be able to fix everything, everytime 100% perfectly with avisynth only

Good luck