getting extra dupes in TFM/Tdecimate

[spiritgumm]

This video (or film) has 2:3 pulldown (it's 29.97) as shown by stepping thru it with separatefields. However, the 23.976 result after using TFM/Tdecimate shows duplicate frames. It's not just at scene changes. I tried all the modes/pp settings for TFM, and the one that worked for this clip (mode=7, pp=7) produced a frame drop later in the video (not in this clip) which was normally deinterlaced or field matched. I'm wondering if a threshold setting would work. I'm assuming that using another command in Tdecimate (I tried sundry) to remove extra dupes would also reduce the framerate, so that doesn't seem like the correct approach. Seems like it's TFM not outputting dupes in a correct order for Tdecimate.

[jagabo]

Automatic gain is causing some fields to be mis-identified at scene changes. A manual IVTC will work for the clip:

Mpeg2Source("VTS_02 clip.d2v", CPU2="xxxxxx", Info=3) 
Separatefields()
SelectEvery(10, -1,0, 1,2, 4,5, 6,7)
Weave()
vInverse() # remove some false residual combing from compression artifacts and per-field brightness changes

You'll have to watch out for phase changes with the whole video.

[spiritgumm]

Thanks but I need a more general or automatic solution, not something just for this 8 second clip. I have several videos like this, and I'm not going to step thru them frame by frame.

[Sharc]

You may want to try Telecide from Donald Graft's decomb.dll or DGTelecide from DGdecodeNV.dll if you have an NVIDIA GPU as an alternative. Sometimes it gives better results.

[spiritgumm]

You may want to try Telecide from Donald Graft's decomb.dll or DGTelecide from DGdecodeNV.dll if you have an NVIDIA GPU as an alternative. Sometimes it gives better results.

Telecide eliminated the dupes (that I noticed) in the clip so I'm hoping it will be good for complete video and others. Thank you, it didn't occur to me to try another field matching filter.

[jagabo]

This works for the clip:

QTGMC()
TDecimate(Cycle=10, CycleR=6)

[spiritgumm]

This works for the clip:

QTGMC()
TDecimate(Cycle=10, CycleR=6)

That's interesting in that you didn't follow Qtgmc with Selecteven and use plain Tdecimate, which is what I tried. I also tried Bob() instead of Qtgmc but the results in some scenes were poor looking video. But since the video only has pulldown, I thought IVTC should be the first choice and faster than deinterlacing.

[jagabo]

Yes, field matching is usually preferred over QTGMC+Decimation but there are exceptions. Telecide+TDecimate is definitely faster than QTGMC+TDecimate. I'd follow the former with something like vInverse and other filtering to remove remaining comb artifacts (from too much compression and changing levels) and noise reduction.

[spiritgumm]

I added santiag, although it didn't help too much.
I just hope Telecide doesn't cause some issue elsewhere. I was also going to post a different video with a similar problem, after trying various methods including Dedup, and "override" settings in TDecimate. But a preview using Telecide shows it helps with most instances.

[jagabo]

TFM(field=1, , scthresh=16) returns almost the exact same thing as Telecide(). Just slightly different post processing, I think.

[spiritgumm]

TFM(field=1, , scthresh=16) returns almost the exact same thing as Telecide(). Just slightly different post processing, I think.

The video is bottom-field first, and since field and order are set the same in TFM, shouldn't "field=0"?

Inre "phase changes" at bottom of your post #2 above, would DGindex have reported them?

[jagabo]

The video is bottom-field first, and since field and order are set the same in TFM, shouldn't "field=0"?

field in TFM isn't the field order. It's which of the two fields it starts with when matching.

Inre "phase changes" at bottom of your post #2 above, would DGindex have reported them?

No. Phase changes can happen when hard telecined video is edited without syncing with the telecine pattern.

[spiritgumm]

The video is bottom-field first, and since field and order are set the same in TFM, shouldn't "field=0"?

field in TFM isn't the field order. It's which of the two fields it starts with when matching.

In the instructions, they recommend setting "field" the same as the "order" unless you have a problem with frame matching. I just wanted to point out the original video is bottom field first. Regardless, I can see this removes all the dupes in the clip, so I was curious what would happen if I removed the "field" setting. Initially I thought one dupe returned, but stepping thru it several times in the original video, looks like it wasn't necessary. Can you explain why it's necessary.

Telecide worked well in another video from this series, but messed up one frame. Disabling post processing eliminated the issue in that spot, but also makes the video less sharp compared to TFM. There's a "guide" parameter to set it for 2:3 telecine which I woulda thought would fix it, but doesn't.