- Haste makes waste 2003-05-01 14:47:13 And as usual when I make hurried entries, I forgot to write about the point for which I was making the entry in the first place: I got some new codecs and have been experimenting with re-compressing my videos for better quality and file size. The hosting I'm using will probably go away in the not too distant future. I'll probably make fresh copies of my existing videos and upload them to the Carrot. What do people expect in terms of bit rate these days?
-
Cream Etrog
2003-05-01 14:43:42
I got sidetracked upgrading my system, both software and hardware. There was an animation festival here in town and I watched the whole damn thing. Then I bought Dragon Half on DVD and watched that several times. Then I noticed that even though the "clean closing" on the DH disc appears to be 23.976 fps, it nonetheless shows interlacing artifacts, and I went through all of transcode's deinterlacing options trying to figure out which one was most appropriate, and also what was going on in the authoring for that disc, because although I have no immediate use for that footage, I know I'll want it eventually, and more knowledge equals more better.
But, anyway. I've also been working away on my Tori Amos/Hellsing video. Just have one chorus and five or six bars of instrumental stuff to go. -
Making some actual progress
2003-04-15 17:13:32
So. Completed multimedia database class project, handed it in, all on time and stuff. I so rock. Now I have more time to work on my video.
Finished re-doing the parts that had gotten screwed up by the frame rate change. Finished sequencing second verse (i.e. the second verse in the song; I actually haven't started the first verse yet) and did most of one of the musical solo bits. The clustering idea hasn't worked very well, but the chronological part of my visual index has proved to be a big help, though probably not as much help as the text script.
The text script seems to be a really good idea, so maybe I should elaborate on it: what I have is a text file in which I've typed all the lyrics of the song, and marked where the bar boundaries are (the places I've put timeline labels in Cinelerra). As I get ideas for storyline stuff, I type those in, too; then when I actually put a shot in place, I make a line with the words of the song for that shot on one side, and a few words describing the visual on the other. I line those things up in two columns so that when I have a series of shots complete, I can read down one column to read the song lyrics and read down the other for the action.
Some gotchas and kinks: For instrumental bits, instead of recording the lyrics I put a number for the number of beats covered. The musical hook for "Mother" has a very characteristic 3-3-3-3-2-2 beat pattern which I'm using to guide my edits for those parts of the song. It's important to actually transcribe the lyrics from the song because the lyrics booklet that came with the album is a tissue of lies!
Anyway, I've been having fun editing still frames and assembling shots and storylines and so on. Current favourite shot: the screen of Walter's computer as he enters a password and views the Internet snuff film, but instead of the snuff film, I've inserted video from Episode 9 where Boobhanshee is molesting Integra... Total sequenced: roughly 35% -
So, it works
2003-04-12 12:08:00
I recompiled Cinelerra 1.1.5 with the new compiler and tools and it more or less works. Still had to disable the lame, toolate, and mpeg2enc plug-ins. Also, the AVI output is in some way screwed up; I can use transcode to convert Cinelerra's AVI output to MPEG (with the now-working mpeg2enc plugin or the older mpeg plugin) but it has weird noise-like artifacts in it that don't show up when I view the AVI in Cinelerra Dude's own "xmovie" viewer. I'm pretty sure it's not transcode's fault, because it reads and writes other files in those formats with no problem.
I have to re-do all the sequencing I'd done so far, because the frame rate change threw my old XML file off. Fortunately, that is easy to do with the aid of my new visual index.
On the visual index: it's still not perfect, but it's good enough to be useful. Fixing the frame rate and a few bugs in the indexing script brought it within a few minutes over the length of the input footage. Then I discovered that I had accidentally forgotten to index several chapters of video, so throwing those in brought the error to less than one minute. Then I found that the last chapter of every disc, containing a few seconds advertising the next disc, has some kind of glitch in it; transcode stops reading the chapter at the glitch, whereas Cinelerra picks up again (very odd - it shows a single file as two separate clips when you load it), so Cinelerra was seeing a few seconds that transcode didn't. I removed the second clip from each of those chapters, then I applied an empirically-determined fudge factor of 0.1%, and the final result is an index that agrees with the Cinelerra time scale to within about one second per hour, which is close enough.
Unfortunately, I'm in crunch mode for school with a 30-page paper due Monday of which I have only written about half; so it may be a while before I can really play with this. -
Compiler purgatory
2003-04-11 12:27:43
So a bunch of people told me instead of using the "mpeg" plugin to transcode I should use the "mpeg2enc" plugin, which is a front-end to the mjpegtools encoder. A few people told me I should punt on transcode entirely and just use mjpegtools directly. So either way, that means compiling mjpegtools.
Downloaded it, tried to compile, it wouldn't, because it wanted avifile.
Downloaded avifile, tried to compile it, it wouldn't, because it included ffmpeg and my compiler couldn't handle the MMX stuff in ffmpeg.
Attempted to disable MMX support. Discovered that although there is a --disable-x86opt option, there is also a hardcoded assignment to "yes" so that although the option of disabling MMX is implemented, it never actually gets turned on. Removed hardcoded assignment.
Package refused to compile because it couldn't find a Windows installation on my machine. Attempted to disable Windows support. Discovered that although there is a --disable-win32 option, there is also a hardcoded yadda yadda yadda the point being that they won't let you disable Windows support either, not even with the option that is documented as doing that. Hacked around this limitation too.
Build still didn't work.
Decided, fuck this, I have an MMX machine anyway, I ought to be able to compile MMX code and have it really work. I'm sick of lobotomizing everything. Downloaded new version of C compiler. Used old compiler to compile new compiler. It didn't work - you must already have the new compiler in order to compile the new compiler, apparently.
Did more research, discovered that it was actually my binutils that was the problem for compiling the compiler. Downloaded new binutils. Compiled it with old compiler and old binutils and it actually worked. Used new binutils and old compiler to compile new compiler. That worked too; it was not difficult, but was painful, because it meant compiling not only the C compiler but also the C++, Objective-C, FORTRAN, and Java compilers, and their associated libraries. Didn't dare disable any of those because I know the day will come when I want to actually compile code in each of those languages. Worst part was waiting for the Java SQL library to compile, because of course Sun had to make that part of the core language library and I know I'm not going to use it by the time it's obsolete, and it took forever to compile. I mean, why should I compile Java SQL libraries just to transcode MPEG video? Drew the line at Ada, because the Ada compiler is partly written in Ada and so you must already have an Ada compiler to compile the Ada compiler. Read some of the Ada compiler's source code just for laughs, because I've never seen live Ada code before. It appears to be five parts Pascal, two parts COBOL, one part FORTRAN.
Used new compiler and binutils to compile standalone ffmpeg. It worked, MMX and all. Used them to compile avifile, then mjpegtools. Those worked too. Attempted to compile transcode. It refused to work with avifile directly. Disabled avifile support in transcode - since it still works with mjpegtools, and mjpegtools is a front-end to avifile (which is in turn a front-end to ffmpeg), and transcode also supports ffmpeg directly, it didn't seem important to have direct avifile support in transcode. Found that transcode still wouldn't work, due to include file path silliness with freetype2. Attempted to disable freetype2 support. Found that although configuration script has option to disable freetype2 support, that option is ignored. WHAT IS IT WITH THESE PEOPLE AND IGNORING CONFIG OPTIONS?? Hacked on freetype2 until I got it to actually work. Compiled transcode successfully.
Now... I have (meanwhile) completed my long drawn-out re-rip and de-interlace of Hellsing. Unfortunately, this means that the Cinelerra XML file I've been building, with the old incorrect-frame-rate rip, will have to be almost entirely re-done. The good side is that now I know just how to re-do it. Furthermore, I've had pretty good results with my HTML indexing experiments and I think I now have a pretty HTML index that will actually correspond to Cinelerra's time measurements, making the XML reconstruction even one more step easier.
And since I'm on such a roll with this compiling, maybe now I'll be able to compile Cinelerra 1.1.5 or whatever the newest version is by now, since it looks like it was the compiler MMX support that was the real sticking point there.
JOURNAL: koronoru (Kevin Oronoru)
Current server time: Dec 25, 2024 19:24:57