Log in

View Full Version : Problems with Decomb


yingx2
26th June 2003, 16:39
I cannot correctly ivtc this clip
http://myweb.hinet.net/home4/yingx2/test.avi

No matter how hard I tried, it seemd that I just couldn't recover its original frame rate and progressive frames with Decomb.

When I used Decimate(cycle=5), the result looked jumpy.
Using Decimate(mode=1) instead, I could not get a 24fps encode.

And I can see fields being blended in cases where they should've been properly ivtced. (Bad source?)

What would you suggest?
Thanks in advance.

Guest
26th June 2003, 17:23
Your avi bombs in Windows media player and everything else I tried. What codec did you use and did you test the uploaded file?

yingx2
26th June 2003, 18:29
It was encoded with Picvideo MJPEG Codec.

I tested it after seeing your post, without noticing big problems, except the fact that the first frame appeared to be a broken one - don't know why.

I uploaded the same clip using xvid at quant 2. Maybe you are still willing to try this one.
http://myweb.hinet.net/home1/joefox/test2.avi

Guest
26th June 2003, 19:40
OK, I opened it with the VirtualDub internal MJPEG decoder.

You have what would be 29.97fps progressive if you could get rid of the blended fields.

Strangely enough, I was alerted by one of my users to an ingenious way to use Decomb 4 to get rid of blends:

Telecide(post=false)
Telecide(reverse=true,post=true)

The result is pretty good 29.97fps progressive. No need to decimate it.

Alternatively, you have to just treat it as interlaced video and deinterlace it.

bilu
27th June 2003, 00:42
Originally posted by neuron2
Strangely enough, I was alerted by one of my users to an ingenious way to use Decomb 4 to get rid of blends:

Telecide(post=false)
Telecide(reverse=true,post=true)

The result is pretty good 29.97fps progressive. No need to decimate it.

Alternatively, you have to just treat it as interlaced video and deinterlace it.

Geez... :eek:
How does that work?
And why can't it work in Decomb 5?
Is it better that Deblend?


I'm still shocked :eek:


Bilu

Guest
27th June 2003, 02:19
Originally posted by bilu
Geez... :eek:
How does that work?
And why can't it work in Decomb 5?
Is it better that Deblend?
I'm still shocked :eek:
Bilu About Decomb 5... It already can handle it nicely with:

Telecide(order=1,triple=true,post=2,vthresh=45)

I didn't point that out because Decomb 5 hasn't escaped from the Development forum yet. :)

The problem with clips like this is that there are field blends scattered sporadically in the top and bottom fields. If on each frame we can match using the best field and not the blended one, we get rid of a lot of the blending. In Decomb 5, it is done directly with triple=true. In Decomb 4, the silly trick above allows matching to be done on both the top and bottom fields. It's not fully correct but does make better output than matching on only one field.

A proper deblending solution would be better. But I'm not convinced that such a thing can exist.

yingx2
27th June 2003, 04:41
Thank you for your magic script and your explanation:)

The result DOES look more progressive than my previous effort. But I can see duplicate frames distributed throughout the whole clip without a mathematical pattern.

f5 = f6
f11 = f12
f18 = f19
f24 = f25
f30 = f31
f36 = f37
f41 = f42
f47 = f48

Is it "29.97fps progressive" ? Do I have to decimate it?

Guest
27th June 2003, 04:59
As I explain in my website journal, doing three-way matching can cause spurious duplicate frames to be emitted. It can also cause frames to be dropped. You have to choose your lesser evil from the options available.

You can certainly remove duplicates by decimating. When the duplicates are random like this, however, you might be better off with MultiDecimate() rather than Decimate().

yingx2
6th July 2003, 17:10
Hi, I'm back.

just finished encoding the whole movie using your suggested script without decimating it. I know that I lose a good compressibility gain by leaving it at 30fps, but the result still looks better than what I could expect.

Thanks a lot.

scharfis_brain
6th July 2003, 17:49
this clips seems to be a progressive 25fps-PAL-video which ahs been converted to ntsc using fieldblending.

a short test with

separatefields()
repal()

helped against the blends, but the motion was a little bit jittery.