No idea how to decimate this capture.

[Yufi]

I've recently been trying to encode a recent capture of Gone with the Wind off of TNT. However, it has quite the odd telecining pattern in it. I've been trying to figure this out, but I simply can't. Since this is such an old movie, I'm not sure what was done to it.

I loaded up a .avs script with just the .d2v being loaded and Telecide(), and then tried to write down the pattern of duplicates. After 7 seconds I gave up, as I could see no visible pattern. I'll paste all that I wrote down below, the basic layout I wrote is <non-duplicate>:<duplicate>:<non-duplicate>:<duplicate>, so 4:1:3:1 is four non-duplicates, followed by a duplicate of the fourth frame, followed by three non-duplicates, followed by a duplicate of the third frame, etc etc.

4:1:4:1:2:1:4:1:2:1:4:1:2:1:4:1:4:1:2:1:4:1:2:1:4:1:2:1:4:1:3:1:3:1:4:1:2:1:4:1:2:1:4:1:4:1:2:1:4:1:2:1:
4:1:2:1:4:1:3:1:3:1:4:1:2:1:4:1:2:1:4:1:2:1:3:1:4:1:2:1:4:1:2:1:4:1:3:1:3:1:4:1:2:1:4:1:2:1:4:1:2:1:4:1:
3:1:4:1:4:1:2:1:3:1

I cut each line off, just imagine as if the above is one big long line instead of three shorter lines. Now, maybe it's just me, but I don't see a pattern above at all. IIRC, when I saved the video with DGIndex, it displayed it as being 100% Video.

There has to be some way to fix this video (maybe there is a pattern, it's just 30 seconds long?) besides just deinterlacing and calling it done. Am I doomed to have to keep looking for a 10+ second pattern in this? Or is there some easy way to figure it out?

[Mug Funky]

29.97 * (82-26)/82 ~= 20.47 fps

i could be wrong though - i only counted the first row...

try not running telecide, doing a separatefields or bob, and counting how many fields each frame gets. i think it might average at 3 fields per frame. also it'll tell us whether the source is blended or nicely done (i think it might be nicely done or you'll have seen interlacing and blending after telecide).

[Yufi]

ok, I tried with SeparateFields(), and here's how it looks, this with the same <non-duplicate>:<duplicate> pattern. Obviously there's more "duplicates" since the fields are separated:
1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:2 1:2 1:1 1:2 1:1 1:2
1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:1 1:2 1:1 1:2 1:1 1:2
1:1 1:2 1:2 1:2 1:1 1:2 1:1 1:2 1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:2 1:2 1:1 1:2 1:1 1:2
1:1 1:3 1:1 1:2 1:1 1:2 1:1 1:2 1:2 1:1

Again, I gave up about 2 seconds (145 "frames" in vdub) in because I was sick of counting and occasionally losing track and having to back count.
Anyway, a couple times it came close to being a pattern, but nothing actual.

[plugh]

Hope people won't mind me reviving this topic (came up in a google search), but I'm also having difficulties deinterlacing a TNT-HD capture of this movie and am looking for assistance...

Using Seperate Fields, I've sort of found a cycle ("|"=new frame, "-"=more fields)

|-|--|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|--|-|--- (3 times)
|-|--|-|--|-|--|--|--|-|--|-|--|-|--|--|--|-|--|-|--|-|--- (once)
|-|--|-|--|-|--|--|--|-|--|-|--|-|---|-|--|-|--|-|--|-|--- (3 times)
|-|--|-|--|-|--|--|--|-|--|-|--|-|---|-|--|-|--|-|--|--|-- (twice)
|-|--|-|--|-|---|-|--|-|--|-|--|-|---|-|--|-|--|-|--|--|-- (twice)
|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|--|-|--|--|-- (once)
|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|--|-|---|-|-- (5 times)
|-|--|-|--|--|--|-|--|-|--|-|---|-|--|-|--|-|--|-|---|-|-- (3 times)
?*2
|-|--|-|---|-|--|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|-- (3 times)
|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|--|-|---|-|--|-|-- (5 times)
|-|--|--|--|-|--|-|--|-|---|-|--|-|--|-|--|-|---|-|--|-|-- (3 times)
|-|--|--|--|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|-- (twice)
|-|---|-|--|-|--|-|--|-|---|-|--|-|--|-|--|--|--|-|--|-|-- (twice?)
?

the 'gap' is a section with to little movement for me to determine if the pattern changed or not (ditto for very last line).

Each line above is _58_ fields. Counting film-frame starts, I see 22 movie frames per each 58 fields (29 video frames). Noting the three-field repeat ("|---") occurrances, and in particular how the pattern changes with respect to them, it is always a matter of introducing the next film frame slightly later.

Another illustration, perhaps clearer - in addition to Seperate Fields, I add a Select Even and reexamine a portion of the above
||-||||-|||-|||-||||-||-||||- (twice)
||-||||-||-||||-||||-||-||||- (3 times)
||-||||-||-||||-|||-|||-||||- (6 times)
||-||||-||-||||-||-||||-||||- (twice)
||-||||-||-||||-||-||||-|||-| (6 times)
||-||||-||-||||-||-||||-||-|| (3 times)

Again, I see 22 film frames per 29 video frames, with the pattern changes corresponding to occasionally introducing a new frame slightly later.

Has anyone ever seen anything like this before?

As a first pass calc, I come up with a film frame rate of
(22 frames) per (29/29.97 seconds) = 22.736 fps
But given the way the pattern changes, the actual film frame rate is probably slightly less than that...

WTF!?

[scharfis_brain]

upload an unprocessed sample somewhere.
I cannot guess patterns and framerates from others people notations...

[plugh]

http://www.mytempdir.com/1287326

approx 50MB snipped from one of the ts files

thanks!

[plugh]

Been poking at this some more...

To start off, none the net research I've done has indicated that this film was shot at an oddball frame rate. To the contrary, what info I have found is consistant with 24fps. And in that context, whereas various references indicate the movie is 222 minutes long (not including the overture etc), this capture, after commercial removal, is 232 minutes long. So everything seems to be indicating 'they' slowed it down ~5% for the broadcast, giving the oddball film frame rate.

Even though the exact pattern varies, every place I've checked I keep coming up with 7 dups per 29 video frames. Noting that 22/29 = .7586, I tried
seperatefields()
selecteven()
decimate(cycle=8)
decimate(cycle=7)
which gives me 6 out of 8 = .7500
This did eliminate the dups, but also about every four seconds also eliminates a 'good' frame, which is visible during pans.

I'm now giving Tdecimate a try, specifying 7 of 29. Seems to be about three times slower, but it also seems to be doing the "right" thing so far. May not be _exactly_ the correct rate, but if not the places with 8 dups in 29 have to be pretty far apart...

Next up - figure out how to properly deinterlace it... Hmmm

|- |- -| -| -- |- |- -| -| -- -| -| -- |- |- -| -| -- |- -| -- |- |- -| -| -- |- |- --
Aa Bb Bc Cd Dd Ee Ff Fg Gh Hh Hi Ij Jj Kk Ll Lm Mn Nn Oo Op Pp Qq Rr Rs St Tt Uu Vv Vv
needs to be converted to
Aa Bb Cc Dd Dd Ee Ff Gg Hh Hh Ii Jj Jj Kk Ll Mm Nn Nn Oo Pp Pp Qq Rr Ss Tt Tt Uu Vv Vv
which look like it'll have to do field matching both forwards and backwards...

[ChiDragon]

Deinterlace? Just use TFM or Decomb to field match before the decimation. And yeah, TNT (at least TNTHD) does seem to alter frame rates all over the place before applying their own telecine.

[scharfis_brain]

@plugh:
I found that the real framerate is somehow 22.6 to your 22.736 fps.

I found that

crop(240,0,-240,0)
tfm(pp=0)
tdecimate(mode=2 , rate=22.736)

is the easiest way to solve this problem

[plugh]

Thanks scharifis - nice to have the confirmation.
I said I thought it was slightly less than the 22.736 I calc'd.
For future reference, how did you come up with the 22.6 figure?

I wonder if they pitch correct the audio when they slow it down, or just play it 'off-key'...

[scharfis_brain]

first, I looked at the video using plain bob().
after I discovered, that there were no orphaned fields (film-frames that last only for one field) I used
tfm(pp=0)
subtract(last, last.deleteframe(0))

this showed a completely grey image if there was a dupe. also this function sequence returns 29.97 fps.

so I counted the grey-only frames in a portion of the sample. since all grey-only frames correnspond to all dupes, the resulting math is simple. so I concluded that the real framerate is somewhere between 22.5 and 22.8 fps.

also tdecimate isn't really happy with these looooong cycles, so it inserts some dupes now and then...

btw.: what is your desired format after IVTCing this source?

[plugh]

Ah... so your "22.6" was an estimate... thought perhaps you were able to deduce the 'actual' from the pattern changes.

I'm sure there is an integer math way to get the residual from those occasional frame-start shifts in the pattern, but I walked through a couple thousand fields and couldn't find a full cycle...
btw.: what is your desired format after IVTCing this source?

Not sure what you're asking - I want color corrected YV12 to feed to VDM to produce an Xvid avi...

[scharfis_brain]

unfortunately I didn't found a cycling constant pattern.

if you are going for xvid avi you just could leave the odd framerate as it is.
this would also avoid audio transcoding.

[plugh]

Yeah, I probably will leave it as-is, was just wondering...

I've never attempted 'speed shifting' a movie before, but I'm not sure it will be worth the effort for this cap - tnt logos and various other inserts popping up in it :(

Besides, I've already expended my 'learning new things' quota on this conversion ;)

Thanks for the assist...