Try:
Code:TFM(mode=0, pp=0) vInverse() # remove a little residual combing on a few frames with very little motion TDecimate()
+ Reply to Thread
Results 31 to 51 of 51
-
-
thanks, jagabo. I forgot to mention I'm cropping this for MKV. and the miceteeth are prominent in the larger picture.
crop(26,112,-24,-114) -
There are no interlace comb artifacts left after TFM() and vInverse(). There is still a lot of horizontal jitter from time base errors. Is that what you're looking to fix?
-
ugh, I guess so.
When I did a different sample with the line below, I get fine miceteeth on the edges of everything. It looks better than IVTCing.
QTGMC(preset="fast").srestore(23.976) -
I was about to post a sample when the server disconnected (probably weather-related).
I get miceteeth in edges like the nurse's uniform. You probably only need the first and last script lines.
Then again, maybe the miceteeth are more prominent after it's cleaned up so you do need a cleaner.
QTGMC(preset="fast").srestore(23.976)
Greyscale()
MCTEmporalDenoise(settings="medium")
smoothLevels(10,1.1,255,0,255)
crop(26,112,-24,-114)Last edited by spiritgumm; 25th Jan 2015 at 09:55. Reason: script cleaner to better see miceteeth
-
I guess you're complaining about the line jitter from horizontal time base errors. I'd do something like this:
Code:crop(24,112,-24,-112) TFM(mode=0, pp=0) vInverse() TDecimate() QTGMC(preset="fast", InputType=1, EZDenoise=2.0) Greyscale() smoothLevels(10,1.1,255,0,255)
Last edited by jagabo; 25th Jan 2015 at 10:26.
-
I ran both (with cleaner in mine) and see the opposite - more aliasing on the guys shirt in your script. However, the bed (between nurse and guy) is bouncy in mine. And there's miceteeth on the edge of the nurse in both, noticeable if you fullscreen video playback.
I see adding the progressive indicator to QTGMC(InputType=1) gets rid of the bouncing bed issue in my script. Adding MCTEmporalDenoise to your script improves it, I think. But as you say, very slow.Last edited by spiritgumm; 25th Jan 2015 at 13:14. Reason: inputtype=1 for progressive video
-
I don't know what you're seeing but this is typical in what I get:
Frame from the source M2V:
Result of my script (TFM) and yours (QTGMC):
TFM keeps almost oll of the horizontal stripes in the man's shirt and doesn't produce any strobing on the bed. QTGMC (from your script in post #36) loses many of the stripes and the lines in the bed are strobing. You can't see the strobing in a single frame but you can see that the details on the bed are different in QTGMC. -
-
ah, you're right about the combing, I thought my media player was just playing the sample badly.
I ended up using your script with the denoiser to reduce the miceteeth - noticeable along the nurse's uniform if you play it full screen. At least that's how it looks to me.
I'm perplexed by the filter order, and the TFM PP setting. Of the many permutations, I first started with TFM(pp=0).tdecimate, but the aliasing (caused by time base errors?) made it look terrible.
Why does telling TTM not to look for combing help?
Why crop first?
I actually wanted to crop more than in your script - is there a reason you changed the left crop value?
Are all the fields picked by the time QTGMC is called, so it's only function is cleaning up the video? -
mode=0 is telling the filter only to look at the immediately adjacent fields for a match. If you don't use that option it may look at further away fields to find a match. Time base jitter or flickering brightness may confuse the filter into selecting the wrong field.
pp=0 turns off the post processing filter. If there isn't a perfect complementary field for the current field there may be some comb artifacts with whatever field is chosen to complete the frame. The post processor blend deinterlaces those artifacts by default.
To speed up the processing of later filters.
Later filters didn't work with the frame size left by your original crop.
Yes. -
Good script, jagabo. I somehow overlooked the pp=o factor. The script kept me up half the night with this POS sample2 video. Wonder what the maker used for that recording, looks like a horrible DVD/VHS combo.
I don't know why I bother taking up horrible video like that sample. Must have a bent gene thread somewhere. You can fiddle with it forever and come back with even less. Not worth it, IMO.- My sister Ann's brother -
Oh, by the way, the post processor of TFM() sometimes mistakes true details for comb artifacts and messes them up. That's the main reason I used pp=0 it in this script.
-
-
Run the following script on your sample:
Code:Mpeg2Source("sample2.d2v", Info=3) StackHorizontal(TFM(mode=0), TFM(mode=0, pp=0))
-
The distortion and lack of sharp detail in my sample means that TFM wont be matching perfect fields, and will see combing, correct? That's what I thought was implied by the first explanation for disabling PP. When you said another reason was that TFM might mistake true details for combing, I was thinking you might mean something else - such as horizontal or vertical objects. I'm not sure what "true details" means in my sample.
Btw, I tried your script for the movie with my Crop settings and it worked. Adding MCTEmporalDenoise required the frame size to be cropped to a certain dimension. -
-
and I would imagine the lines in the shirt. Makes sense.
So would you say more often than not, (mode=0,pp=0) are more appropriate than the defaults? -