still interlaced frames left

[subcool]

I'v made a .avi from a dvd, most frames are good but there are a few frames are still a bit interlaced. See the link for a screenshot
slightly interlaced frame (http://arienai.com/frame%20480.png) (look at the lights in the bottom left corner)

The source is 29.97 fps, NTSC.

This is what im using now.

MPEG2Source("PoT.d2v").Crop(6, 0, -6, 0).BicubicResize(712, 480, 0, 0.75)
tfm(d2v="PoT.d2v",pp=7,mode=2)
tdecimate()

Anyone know of a way to get rid of the last few interlaced frames?

[slk001]

This might have been in the original. It doesn't otherwise look like it is interlaced.

[tritical]

First, I would suggest doing the crop/resize after tfm()/tdecimate() for a couple reasons. 1.) Even though the resize is horizontal only and not very large it will change things slightly 2.) The framesize will no longer be mod 16 width and tdecimate runs about 8-10 times faster if width/height are mod 16 (if you have the latest version, v0.9.9.1).

There are a few ways you could increase the sensitivity of the combed frame detection. However, I would first suggest commenting out the tdecimate() line, and then adding "display=true" into the tfm() line so you end up with:

MPEG2Source("PoT.d2v")
tfm(d2v="PoT.d2v",pp=7,mode=2,display=true)
#tdecimate()
#Crop(6, 0, -6, 0)
#BicubicResize(712, 480, 0, 0.75)

On the upper left of each frame there will now be some writing, and on the last line there will be "MIC = #". When this value is larger then MI tfm declares the frame combed. By default MI is set to 85 in the latest versions. If the combed frames that are being missed are already close to this value (within 5 or 10) you can simply set MI lower to have them detected. Alternatively, you can try one of 3 things to increase the MIC value of the combed frames:

1.) Use "chroma=true". This will make tfm count chroma combing as well as luma. By default only luma considered. This might help in your case with the lights.

2.) Decrease "cthresh". By default cthresh is set to 10, try lowering it to 7 or 8. This will make tfm more sensitive to combing, but will also give more false detections on the pixel level.

3.) Adjust the size of the window used during combed frame detection. By default a 16x16 window is used. This is a good value when trying to detect badly combed frames and having as few false detections as possible. However, sometimes it isn't sensitive enough to catch very small, separated, areas such as in the picture you put up. You could try setting "blockx=8,blocky=8". This will make tfm use an 8x8 window. MI must then be adjusted appropriately as well, because with an 8x8 window the largest MIC can be is 64. A value for MI between 20-30 would probably be best. A window size of 4x8 (blockx=4,blocky=8), with an MI value of 10-15, would be able to catch even smaller combs. Of course, the smaller the window size the less resistant it is to noise/high frequency detail.

Hope that helped. Of course with my filters there are always at least 500 parameters to change :p.

[subcool]

thanks, i'll try out your suggestions once i get home from work, tritical. hope they'll work ^_^

[eb]

... to help in viewing

http://img177.echo.cx/img177/9702/interlaced6eb.jpg (http://magnification]magnification[/URL]

[subcool]

When i try this AVS, the combed files are properly dealt with =)
thanks for the help Tritical and for making such a great plugin.

MPEG2Source("test.d2v")
tfm(d2v="test.d2v",pp=7,mode=2,chroma=true,cthresh=7)
#tfm(d2v="test.d2v",pp=7,mode=2,chroma=true,cthresh=7,blockx=8,blocky=8,MI=25)
tdecimate()
Crop(6, 0, -6, 0).BicubicResize(640, 480, 0, 0.75)

When i try the commented line btw, it decombs good frames too lol...

[Revgen]

Try the tryWeave=true parameter in tdinterlace and see if it looks better. AFAIK this paramerer allows tdint to automaticaly switch to per field deinterlacing in case per pixel motion deinterlacing doesn't create a desired result.

[Wilbert]

AFAIK this paramerer allows tdint to automaticaly switch to per field deinterlacing
What's that?

[subcool]

Hmm it seems there are still some combed frames in the video and at some point there's rainbowing in the background.

i must say, this source is really annoying. mainly cause of all the lights moving about and fast motion (dancing)

[Guest]

Can you post a link to a short unprocessed source clip?

[tritical]

Try the tryWeave=true parameter in tdinterlace and see if it looks better. AFAIK this paramerer allows tdint to automaticaly switch to per field deinterlacing in case per pixel motion deinterlacing doesn't create a desired result. Almost, what it really does is switches from per-pixel to per-field motion adaptivity. Basically, all it will do with tryweave=true is test the matches that tfm would do in mode=0, and if the best one is not combed then it uses the weaved frame instead of deinterlacing the input frame. Anyways, the latest versions of tdeint use exactly the same combed frame detection method as tfm with exactly the same 5 settings (MI, cthresh, blockx, blocky, chroma) but slightly different defaults.