Telecide questions 3:2 pulldown

[pdottz]

so after actually seeing that none of my toon caps are interlaced i figured out that they are 3 frames prog, then 2 interlaced.

this is the line i now use in my script to fix this.

Telecide(order=1,guide=1,post=0)


this fixes it and does a nice job.

however now i get small 'bumps' on those pulled frames.

for instance, this is a 3:2 frame

||
||

now after the 3:2 pulldown it looks like this


|.
|.

the dots represent the lines im talking about. i'll post a pic later when i get home.

can anyone recommend something to fix this? or something i'm doing wrong?

my caps are NTSC 29.97

thanks for any help :)

[foxyshadis]

Sure, try post=2, it's there to get rid of leftover combing artfacts even if it's a bit slower. You can tweak v/dthresh to see if you can get better results once you enable that.

Once you've exhausted all of that, if it still isn't acceptible you might need to switch to tfm (and maybe post a sample for someone to help with a better script), but I think this will help the quality a lot.

[Guest]

It's not clear to me that he is talking about residual combing.

@pdottz

Please post a screenshot, or better yet, post a short unprocessed source clip.

Also, try without the guide parameter; sometimes it doesn't work good with anime.

[pdottz]

sorry it took so long, i was just waiting for a good cap that showed my problem.

here is a frame, that had the 2 frames interlaced on another section of the image. NOT on the step in the pic. this was caused by using

Telecide(order=1,guide=1,post=2)

or

Telecide(order=1,guide=1,post=0)

http://www.disillusioned-inc.org/dotted_3_2_frame.png

now here is the same unprocessed RAW frame.

http://www.disillusioned-inc.org/clean_3_2_frame.png

now i'm sure the video has the 3:2 affect since i frame by frame it in vdub for about 1000 frames counting.

thanks for any help :)

[Guest]

It's probably just a failure of field matching for this frame. If you post a link to a source clip I'll tell you if any parameter tweaks can help.

Generally, occasional field match failures can occur, but they are usually imperceptible when the clip is played at normal speed and sizes.

[pdottz]

neuron.
thanks for the help. some of the problem was actually the post=2 parameter in Telecide...(i changed it to post=0) post=0 actually gives me the little dots on the lines.

i still get some weirdness, but its far and few in between. i actually view all my encodes on the tv using my xbox, however since the flaws are visible on my pc monitor, i figured they'd be visible on an HDTV when i get one....i'll post the sample clips in about 2 hrs since they're uploading to my website right now. maybe you can still help me tweak it up a bit.

again, thanks for the help neuron and foxyshadis.

[pdottz]

neuron.

thanks for the help. here are some sample clips.

16990_raw.rar is an unfiltered pure cap. mjpeg. 16990 referes to the frame i started stepping from so it's not important.

16990_process.rar is the same cap but was processed using this script...
everything in this script has worked fine for me in the past except the telecide. i use to use tomsmocomp up until about a week ago.

LoadPlugin("C:\avisynth\audgraph.dll")
LoadPlugin("C:\avisynth\undot.dll")
LoadPlugin("C:\avisynth\degrainmedian.dll")
AVISource("D:\captures\maincap\grim.avi")
Telecide(order=1,guide=1,post=2)
Crop (4,4,0,0)
DeGrainMedian(limitY=5,limitUV=7,mode=1)
Undot()
Lanczos4Resize(512,384)
ConvertToRGB()
#audiograph(last,2)


http://www.disillusioned-inc.org/16990_processed.rar

http://www.disillusioned-inc.org/16990_raw.rar

[Guest]

Sorry but that clip is poor for analysis. Please provide one with good motion, like a pan or something moving across the screen. Thank you.

[pdottz]

neuron.

now i feel like an idiot for having posted so much and not running more tests. my caps look so much better with post=0 the artifacts left over that i am talking about i think where left over from when i used degrainmedian right after using tomsmocomp in my scripts.

now that i have switched from tomsmocomp to telecide and from fluxsmooth to degrainmedian, everything looks better.

after having used the above posted script with post=0, those little dots are almost non apparent. and i think what is left over is really just dot crawl resulting from a noisy cap that degrainmedian couldn't take care of. however, in case you are still curious, here is a highmotion scene in processed, and raw cap.
http://www.disillusioned-inc.org/swing.rar


again, thanks for all your help neuron. i didn't know a thing as bad field matching existed. i'll search around the forums on the subject in case i ever run into the problem.

[Guest]

There's something really messed up with your source files. The fields are not properly aligned. Even when the fields are matched, everything is combed, even though the fields appear to be from the same picture. It looks as if one field is out of place vertically. It's horrible; you'll never good good quality from that source.

EDIT: Your fields are swapped! I'll post a script to fix that shortly.

[Guest]

I would not use guide=1 on this material. Due to the limited base rate of the animation, there is not enough motion to make it reliable. It might cause false matches in low-motion sections.

Also, since you are converting to RGB anyway, why not use MSmooth to get rid of the grain and the mosquito noise in one fell swoop? Like this:

avisource("swing_raw.avi")
assumetff()
telecide(post=0)
converttorgb()
msmooth(threshold=25,strength=4)