TDeint and TIVTC

[Moitah]

Does anyone else see a bunch of green stuff when using TDeint(type=1) with TDeintv1b3? Source frame, After TDeint(type=1) (Sorry they are a bit blurred from the JPG compression)

[tritical]

Yep, it is a bug in YUY2 type=1 chroma interpolation... forgot to change three lines when ap was added in beta 3. All the YV12 routines and all the other YUY2 routines are ok. I'll try and fix that soon. Thanks for reporting.

[acrespo]

Can I trim video before use TFM with flags= 0 to 2 like this:

mpeg2source("smx.d2v")
Trim(449,32701)++Trim(35087,47956)
TFM(d2v="smx.d2v",mode=3,pp=7,slow=2,chroma=true,flags=3)
TDecimate(mode=1,hybrid=1)

[tritical]

You can, but it wont work correctly . For flags=0 to 2 there can't be any modification of the # of fields or the order of those fields between mpeg2source() and tfm()/tdecimate(). flags=3 as you have now will work ok though. If you want to use flags=0,1, or 2 you would need to move the trim to after tdecimate.

[tritical]

Here is [link removed].

This version adds a hybrid=3 option to tdecimate which is similar to hybrid=1 except that instead of leaving 24p sections untouched and blend converting 30p sections to 24p, hybrid=3 leaves 30p sections untouched and blend converts 24p sections to 30p. This version also adds isse optimizations for more blocksizes for tdecimate's metrics calculations. The last addition is a new parameter called "sco" to TFM which can be used to allow u/b matches only at scenechanges (good for dealing with the case of orphaned fields that only appear at scenechanges due to edits). The only thing left on the todo list for tivtc at this point is optional blending in mode 2 of tdecimate.

[tritical]

Here is [link removed].

Changes:

- SetCacheHints call to diameter instead of radius
- Fixed type=1 YUY2 interpolation routine giving messed up chroma output
   (bug was introduced in v1.0 beta 3)

[Leak]

Originally Posted by tritical

The last addition is a new parameter called "sco"[...]

Sure beats writing "Litigious Bastards"...

Anyhow, I'd have one request for TIVTC - while blending 30FPS sections down to 24 works fine, I've found that some of the CGI car scenes in GITS:SAC were oddly enough rendered in 15FPS, i.e. every frame is shown twice.

The problem here is that TIVTC decimates these scenes since it thinks it's material that had pulldown applied to it, and thus introduces jerkiness. Would it be possible to check if every other frame (or probably 2 out of 3 and 3 out of 4 frames) is a duplicate and in that case treat this scene as video instead of film?

Yeah, I know it's a rather odd case, but it exists nonetheless...

np: Richard Devine - Arc-Acid (Cautella)

[scharfis_brain]

Well, as I asked before:is it possible to code a mode switcher?
(modes: Film, 30p, 60i)

like this pseudo-code for standards conversion:

mpeg2source("NTSC-Hybrid.d2v")

a=bob().convertfps(50) # 60 to 50
b=telecide().changefps(50) # 30 to 50
c=telecide().decimate().changefps(50) #24 to 50
T-Switch(last,a,b,c)
resize(width,576)
assumetff().separatefields().selectevery(4,0,3).weave()

[Leak]

Originally Posted by scharfis_brain

Well, as I asked before:is it possible to code a mode switcher?
(modes: Film, 30p, 60i)

like this pseudo-code for standards conversion:

Wouldn't that have the problem that there could be discontinuities at the points where the mode is switched? I.e. a frame might be repeated since both the film and 30P stream considered it essential at that spot, or a frame might be skipped for similar reasons; which could be prevented by a single filter that handles everything.

np: Jah Wobble - Fly 2 (I Could Have Been A Contender (Disc 1))

[scharfis_brain]

this could maybe be avoided by checking, which of the new rendered frames is closer to the before chosen one...

[tritical]

@Leak
Could you send me a sample from gits:sac to test on? Such cases are indeed not handled correctly with vidDetect=3. You would probably be better off using vidDetect=0 or possibly vidDetect=2, but I doubt it will fix everything.

@scharfis_brain
It is possible of course, but I've already got a lot to do on my existing filters and I have no material to test on that is a mix of 24p/30p/60i. Would you be able to send some samples?

[Leak]

Originally Posted by tritical

@Leak
Could you send me a sample from gits:sac to test on?

Sure... give me a minute or five...

Here it is.

Actually, after taking a closer look at this sequence it's not simply 15 FPS animation, but something quite strange...

Such cases are indeed not handled correctly with vidDetect=3. You would probably be better off using vidDetect=0 or possibly vidDetect=2, but I doubt it will fix everything.

I doubt doing that for maybe 15 seconds of footage (if even) for a 24 minute episode is worth it...

np: Jah Wobble - Fly 2 (I Could Have Been A Contender (Disc 1))

[tritical]

I looked at the clip and to me it looks like plain 12fps animation. After field matching there is a 3 frames (1 new + 2 dups) followed by 2 frames (1 new + 1 dup) pattern and mode = 1 with tdecimate handles it fine. It did have some irregular rff flagging so using d2v="" in tfm with flags = 0, 1 or 2 lead to incorrect decimation in a couple cycles where mode=1 w/o the rff info picked the correct frames to decimate. I went ahead and added in some extra logic to prevent mode=1 using d2v dup info in such cases.

Here is the result (xvid const quant 2) after using the following script:

mpeg2source("C:\gits.d2v")
tfm(d2v="C:\gits.d2v")
tdecimate(mode=1)

gits clip

[Mug Funky]

yeah. GITS SAC sticks to 24fps all through except in the last couple of episodes, and the japanese intro sequence (which comes out slightly jerky in the PAL version due to decimation). i think towards the end of the show some shortcuts were taken in editing, so there's a mix of pure 24p and some 30p. but i can only say this based on the PAL conformed version (which has some full-frame blending in it on the last tape that wasn't on any of the others!).

kinda makes me wish there were a way to get the best of all worlds in standards-conversion. speed up the film bits, blend the 30p and 60i bits to 50i, and encode interlaced.

Originally Posted by tritical

It is possible of course, but I've already got a lot to do on my existing filters and I have no material to test on that is a mix of 24p/30p/60i. Would you be able to send some samples?

the intro to Gantz has all three types of content. anywhere you want it uploaded? i'll grab it tomorrow (at home now...).

[tritical]

@Mug Funky

68.119.245.113
port 17252
upload/upload

If that wont work just pm me.

I've seen the intro sequence for the first gits:sac r1 dvd and it has 30p, but I think most of the actual ep is 24, if not all, though I couldn't guarentee it.

[scharfis_brain]

@MugFunky: DEFT converters can do this. (I don't know what their price is)
but they are still inperfect, because they are deinterlacing and blending the 30p to PAL
.
I prefer fieldmatched 30p, which has became telecined (changefps) to PAL.

[tritical]

[link removed]

Changes:

TDecimate:

   + mode=1 w/ d2v info and hybrid=3 decimation improvements

[B.F.]

New DGMPGDec is out.
d2v opinion don't work now.

[tritical]

Support will be added in the next release, but I am working on adding some other features so it won't be out immediately. For now, to use the d2v option with v11 d2v files just make a copy of the d2v file and in the copy change the very first line from:

DGIndexProjectFile11

to

DGIndexProjectFile10

and use that copy as the file for the d2v argument in tfm(). The only change in v10 to v11 was removing a number before the path entries and that wont make any difference to tfm's d2v parsing/reading.

[Mug Funky]

Originally Posted by tritical

I've seen the intro sequence for the first gits:sac r1 dvd and it has 30p, but I think most of the actual ep is 24, if not all, though I couldn't guarentee it.

all of it except the last 2 episodes. don't know why it went that way (deadlines in production?).