***Formula to guess the Decimate cycle***

[bilu]

Have a look:

http://forum.doom9.org/showthread.php?s=&postid=431541#post431541

This could be very neat for weird patterns like silent movies :)


c= combed frame number
fc = former combed frame number
bfc = before former frame number

if (fc=c-1 OR fc=bfc+1) then cycle=c-bfc

Example 1: Telecine (cycle=5)
=============================

ABBCDEFFGH
ABCDDEFGHH
ppccpppccp (p=progressive c=combed)

frames before combed:
ccppp fc=bfc+1
cpppc fc=c-1

cycle=c-bfc=5

Example 2: PAL2NTSC (cycle=6)
=============================
aka 2:2:3:2:3 pulldown

ABCCDEFGHHIJ
ABCDEEFGHIJJ
pppccppppccp (p=progressive c=combed)

frames before combed:
ccpppp fc=bfc+1
cppppc fc=c-1

cycle=c-bfc=6


New version: IF (fc=c-1 OR fc=bfc+1) THEN cycle=c-bfc ELSE cycle=c-fc (version able to deal with cycle=3)


Bilu

[bilu]

Last version doesn't work on patterns like this:
3000130001

AABCDE -> ABCDEE
ABCDEE ABCDEE

AABCDEFFGHIJ
ABCDEEFGHIJJ
pccccppccccp

c ppccc fc=c-1 AND fc=bfc+1 -> cycle=2
c cppcc fc=c-1 AND fc=bfc+1 -> cycle=2
c ccppc fc=c-1 -> cycle=4
c cccpp -> cycle=3
Working on a new one...


Bilu

[bilu]

Consider these patterns:
23012301 -> cycle 5

ABBCDEFFGH
ABCDDEFGHH
ppccpppccp

30013001 -> cycle 5

AABCDEEFGH
ABCDDEFGHH
pcccppcccp

3131 -> cycle 3

AABCCD
ABBCDD
pcppcp

p=progressive
c=combed
Safest way to define a pattern cycle? The longest distance between progressive or interlaced frames + 1 (if cycle>3) or +0 (cycle=3).

Assuming:
p=progressive
fp=former progressive

c=combed
fc=former combed

pat=(c-fc > p-fp) 1:0
off_c=((pat==1) && (c-fc==3))? 0:1
off_p=((pat==0) && (p-fp==3))? 0:1
cycle=(pat=1)? (c-fc)+off_c:(p-fp)+off_p
Now every pattern, no matter how weird it is, gets detected :)


Bilu

[bilu]

Small correction:
p=progressive
fp=former progressive

c=combed
fc=former combed

pat=(c-fc > p-fp) 1:0
off_c=((pat==1) && (c-fc<4))? 0:1
off_p=((pat==0) && (p-fp<4))? 0:1
cycle=(pat=1)? (c-fc)+off_c:(p-fp)+off_p
Now able to deal with cycle=2 and cycle=1 (duh!).

Pretty nice for stuff like a Neo Genesis clip I have:
A B C
A AC C
AC is a blended field from A and C, B is an orphan field. So its combed at every 2 frames.

The purpose of this stuff is statistical, to help you make decisions over what you can do with weird streams.


Bilu

[Mug Funky]

nice work... i'd like to bump this because i don't have any sources that choke with simple telecide default (good telecine must be a PAL thing :P)

[bilu]

This is mostly useful to:
1) Identify Video and Film parts on hybrid streams;
2) Identify PAL to NTSC conversions made with the 2:2:3:2:3 process;
3) Identify large patterns with silent movies;

As you can see in http://www.doom9.org/ivtc-tut.htm "SILENT FILMS", you have streams with multiple Telecine patterns one over another, and you have to Decimate more than one time. Maybe this can help too in those cases, at least the composed pattern shall repeat itself :rolleyes:


Bilu

[scharfis_brain]

decimate(6) will restore PAL from 2:2:3:2:3 Pulldown, but the resulting
framerate is 24.975 fps instead of 25fps, due to the 30.00 <-> 29.97 issue. This means every 1000th frame of the PAL-Video will be dropped.
a decimate cycle of 2997/497 (6.03018108651) is needed. Unfortunately decimate doesn't support float-cycles...

is there any way to get true 25fps without using change/assumefps afterwards?

[bilu]

I don't know anything else related to 2:2:3:2:3 pulldown beyond Telecide(guide=3).Decimate(6) :confused:

And since you are deconstructing a specific method, you'd be safer this way IMHO.

Bilu