[MPlayer-users] new version of detc filter! (inverse telecine)

[D Richard Felker III]

On Sun, Mar 16, 2003 at 09:44:54PM +0300, Vladimir Mosgalin wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> On Sat, 15 Mar 2003, D Richard Felker III wrote:
> 
> DRFI>Postproc should probably be done afterwards for reliability and
> DRFI>quality. Otherwise, detc might not detect interlacing as well and
> 
> Ok... But there was something about 'pp' should be the first filter in
> chain...

Bleh, yes, I wrote that.... Actually, it's probably a bad idea to use
deblock/dering at all with detc, then...

> DRFI>> But I never reencoded... And I tried it with some svcds, it
> DRFI>> worked great.
> DRFI>
> DRFI>Yes, except it uses tons of CPU time and every 4th frame will be
> 
> Huh? It takes less then 1 percent of cpu time... Denoise3d is the most
> cpu-eating filter, in my opinion - > 10% for 480x480 video.

Ah, you must have a 3 GHz Athlon or something.

> DRFI>shown for 1/15 second (double time) instead of all frames being
> DRFI>shown for 1/24 second. This looks really bad during pans.
> 
> Well, I see that something isn't too good, but at least it's much better
> then original video or deinterlaced video.
> 
> OK, now I tried to reencode video, but something isn't working right. I
> tried -vop detc, -vop detc=dr=1, with -ofps 23.97/23.976/23.98/24, but I
> see duplicate frames messages and sometimes interlaced frames appear in
> resulting file. There aren't too much of them, in average every 22-th
> frame is "duplicate", sometimes there are 200-300 frames in a row
> without duplication, sometimes two following frames are both
> "duplicated". Here is an example (output from "mencoder -vcd 1 -o
> naus.avi -ovc lavc -oac copy -vop detc=dr=1 -ofps 23.97"):
> 
> Pos:   1,7s     56f (147%)  53fps Trem:   0min   0mb  A-V:0,022 [872:128]
> 1 ?????????????(???) ????(?/??)!
> Pos:   2,2s     72f (153%)  56fps Trem:   0min   0mb  A-V:0,028 [760:128]
> 1 ?????????????(???) ????(?/??)!
> Pos:   2,8s     88f (158%)  58fps Trem:   0min   0mb  A-V:0,022 [715:128]
> 
> More of the pattern:
> 30f 56f 72f 88f 94f 111f 117f 138f 154f 159f 174f 184f 189f 199f 204f
> 219f 234f 249f 274f 279f 294f 309f 324f 334f 339f 349f 354f 369f 384f
> 404f 416f 417f 428f 434f 451f 452f 458f 459f 471f 482f 488f 494f 511f
> 518f 524f 544f 554f 574f 584f 589f 604f 614f 634f 649f 659f 664f 709f
> 769f 779f 784f 799f 934f 944f 959f 964f 974f 1193f 1208f 1223f 1230f
> 1238f 1253f 1260f 1268f 1283f 1298f 1313f 1320f 1321f 1322f 1323f 1333f
> 1340f 1348f 1355f 1363f 1378f 1393f 1408f 1423f 1438f 1453f 1468f 1483f
> 1498f 1513f 1543f 1558f 1573f 1580f 1588f 1603f 1743f 1798f 1813f 1820f
> 1828f 1835f
> 
> it seems to be quite irregular... And sometimes it shifts at one frame
> (29 55 71 ...)
> 
> In resulting file in some frames I see patterns which are similar to
> interlacing, but are in color. Is there way to force detc to detect this
> svcd as telecined one?

I'm not sure what's causing it. Maybe upload a segment of the file and
I can test as soon as I get a chance. I'm rather bust these days
though so it might be a week or two before I get around to working
more on it.

Rich