TomsMoComp is bad?

[N1k0s]

Hi, I'm using Gordian Knot 0.35.0 pack with 2.5 AviSynth.
When I save the .AVS file after resizing, the TomsMoComp makes quite ugly deinterlace, although everywhere I read it is the best.
I tried to find the newest TomsMoComp and replace, but the newest was the same what I used before.
Here is 2 test: (The different is only the deinterlace filter)

http://nikosz.netdisk.hu/Lanczos_TomsMoComp.avi
http://nikosz.netdisk.hu/Lanczos_Field-deinterlace.avi

These are with Xvid 1.0.3 codec. I've tried it with DivX 5.2.1 but it gave same result.
Also quite interesting that these two shorts was left the same resolution as the source.

Any idea? Forget using TomsMoComp?
Please help
Nikos

[scharfis_brain]

Why do you try to deinterlace progressive footage?

plus, fielddeinterlace has a progressive frame check, so, all Fielddeint did was NOTHING on your clip.

tomsmocomp is only build to work with real interlaced video, like it comes from video (NOT Film) cameras.

[Guest]

FD roolz! :)

Seriously, though, scharfi, how do you know it is progressive? Those CG segments are sometimes interlaced.

Maybe he has the vertical filter enabled for TMC. That will blur compared to FD.

[scharfis_brain]

FD roolz! :)

cough

Seriously, though, scharfi, how do you know it is progressive?

because the FD-sample is crisp (-> no blending) and doesn't show flicker/stairstepping (-> no interpolation).

So I assume that the source hasn't been modified.
(-> it already has been progressive before)

Tomsmocomp *always* interpolates.
Either with ELA or with its small MC/ME-engine.

But as I said before:
TMC is not suited for partly interlaced stuff.

[Guest]

Well then, if TMC can't be used on hybrid material such as this (let's assume the poster is smart enough not to use a deinterlacer on a clip that is fully progressive), then I repeat: FD roolz! :)

Seriously, why don't more deinterlacers make the progressive frame check?

[Mug Funky]

Seriously, why don't more deinterlacers make the progressive frame check?

could be because loads of DVDs are encoded as interlaced, just to be safe (a few wasted bits at 8.5mbps average is better than mosquitoes in a freak interlaced bit).

it's usually safer using metrics rather than reading a flag.

maybe this is because i see so much field-blended interlaced stuff that could quite easily be progressive if certain companies would have a little more respect for PAL land... okay, rant over.

[Guest]

I was talking about the progressive check in FD. It is a metric check, not a flag check!

[SirCanealot]

Originally posted by Mug Funky
maybe this is because i see so much field-blended interlaced stuff that could quite easily be progressive if certain companies would have a little more respect for PAL land... okay, rant over. [/B]

One wonders why you actually buy that crap :P

[trbarry]

Seriously, why don't more deinterlacers make the progressive frame check?

In the case of TomsMoComp it was because I wanted something that would do the best job on video material where I already knew ahead of time it was interlaced. I'd already written GreedyHMA to handle the mixed automatic stuff but wanted one optimized for the assumption that fields represented time points 1/60 (1/50) second apart. If you are using it on mixed material you can't make that assumption and should be using something else.

I originally intended to have TMC be called by greedy when appropriate but then you (Neuron2) contacted me about writing Decomb and I eventually started using that for IVTC chores instead. ;)

And these days I really don't have the extra time for writing Avisynth filters, so I guess TMC is sort of stuck there. Video only. (and with SE=5)

- Tom

[Mug Funky]

One wonders why you actually buy that crap :P

i don't often buy it. regrettably i'm actually a major part of the problem - i make it.

it's all fine if you watch on an interlaced CRT...

doing it the good way simply costs too much or takes too long, and the bottom line is a DVD authoring house isn't a mastering house.

unfortunately just like with DVD backup, we're often at the mercy of our sources.

[N1k0s]

Originally posted by scharfis_brain
Why do you try to deinterlace progressive footage?

plus, fielddeinterlace has a progressive frame check, so, all Fielddeint did was NOTHING on your clip.

tomsmocomp is only build to work with real interlaced video, like it comes from video (NOT Film) cameras.


OK, sorry, I wasn't clear.
This short clip is from THX1138 DVD (by George Lucas)
The source is interlaced, and I want to deinterlace it with the best deinterlace filter - as I read this is TMC.
So I wonder why TMC makes these errors.
Any more idea?

Thanks
Nikos

[scharfis_brain]

There is no best deinterlacer.

Also there is no best fitting all purpose filter.

Oh wait: Maybe blur(0,1), but that's brute force.

Make a unprocessed VOB File available.
Then we can judge, which filter eliminates the interlacing with the least amount of quality loss.

[Guest]

Originally posted by N1k0s
So I wonder why TMC makes these errors.
Did you have the vertical filter enabled on TMC?

[N1k0s]

Originally posted by neuron2
Did you have the vertical filter enabled on TMC?

I'm not so educated in this theme, only try to use Gordian Knot (ver 0.35.0) as best as possible. But I found some settings in the .AVS file for TMC:

# DEINTERLACING (1)
#FieldDeinterlace()
#FieldDeinterlace(blend=false)
TomsMoComp(1,5,1)

Maybe I need to change this 1,5,1 parameters.

Nikos

[Guest]

Yes, the third parameter controls the vertical filter. You can use this to turn off the filter:

TomsMoComp(1,5,0)

[Mug Funky]

This short clip is from THX1138 DVD (by George Lucas)

hang about... that's shot on film! it should be IVTC'able.

unless the clip is from the special features... then it could be anything.

this is a movie i really must see at some point. i've only seen a clip from it in pop-culture class a couple years ago. good to see it's on DVD now (i wonder if it is made well enough to receive the certification that got it's name from this movie?).

[N1k0s]

Originally posted by neuron2
Yes, the third parameter controls the vertical filter. You can use this to turn off the filter:

TomsMoComp(1,5,0)

So you suggest beyond that to use TomsMoComp whit this parameters above? The original (1,5,1) settings is wrong?
Can you explain when can/must I use vertical filter, and what circumstances is not suggested?

[Guest]

Originally posted by N1k0s
So you suggest beyond that to use TomsMoComp whit this parameters above? The original (1,5,1) settings is wrong?
Can you explain when can/must I use vertical filter, and what circumstances is not suggested? I'm suggesting that the blurriness comes from TMC's vertical filtering. You should try running it without the vertical filtering enabled by using the parameters I gave you. They are not "wrong". It's your decision if you want it or not. Relevant considerations are available in the help file and several threads here.