Log in

View Full Version : How to correctly remove horizontally shifted frames?


bruno321
26th February 2021, 13:21
I posted this on VideoHelp (https://forum.videohelp.com/threads/400835-How-to-correctly-remove-horizontally-shifted-frames#post2611584)some days ago but didn't get any input.

I've this NTSC DVD which displays the following behavior: after QTGMC(), there are some ghosts with purple color and some ghosts with green color. Moreover, there are some not correctly matched frames, as you can see in the picture below. These duplicates follow some pattern, so it should be possible to remove them.

How can I restore this to fluid playback and without the color problems?

https://forum.videohelp.com/attachments/57509-1614069015/since%20snippet000338.png

Here's the snippet (https://forum.videohelp.com/attachments/57508-1614068716/since%20snippet.demuxed.m2v).

StainlessS
26th February 2021, 15:43
displays the following behavior: after QTGMC(),
It displays that behaviour before QTGMC too.
Is NTSC 29.97fps, TFF, with blending fields, and a weird behaviour where every 5th & 6th field,[variable position] upper part of field
seems to be captured a little while before bottom part [where your guys head goes weird, that is a source field].

Dont think anything is gonna help much, but I aint no expert.

EDIT:

VideoFileName ="D:\DVD\since snippet.demuxed.d2v"
MPEG2Source(VideoFileName)
AssumeTFF
SeparateFields
Trim(338,-1) # This seems to be field 334 when play from start (start mid GOP ?)
AssumeframeBased
Spline36Resize(640,480)
Return Last

https://i.postimg.cc/jjW68R4j/since-snippet-demuxed-00.png (https://postimages.org/)

EDIT: D2v flags as 100% video.

bruno321
26th February 2021, 16:26
My impression is that this was converted from PAL video at 49.95fps to NTSC video at 59.94fps via almost dupes. I feel something like QTGMC().TDecimate(0,1,6) should work to come back to 49.95fps, except this tdecimate line is not always decimating what it should. The reason being that the dupe is sometimes not clean, but a messed-up one like in the picture, so tdecimate doesn't get it right.