-
too much fidelity
2002-11-19 13:54:13
So my build of Cinelerra 1.1 doesn't like working with YUVA 16-bit colormodels. DAMNIT.
On Cinelerra's Sourceforge project page, I'm the only one to have mentioned a problem with YUVA 16-bit colormodels. I must have really broken something with the way I compiled it.
Yes, it's possible to really change program behavior if you do something like, say, changing the alignment boundary for jumps by a byte or so, or optimizing generated code in a certain way. For example, the "Play" button doesn't work in Cinelerra if I optimize the code for the Athlon Thunderbird architecture.
Go figure.
I can still work with YUVA 8-bit, and YUV 16-bit, as well as the RGB/RGBA colormodels. However:
(1) I'm directly using the VOBs from the DVDs, and I want to do _NO_ colorspace conversions whatsoever (unlike all you Premiere people who have to do at least one conversion, haha)
(2) I need an alpha channel for the kinds of compositing I do
(3) 16 bits of fidelity is way better than 8 bits of fidelity
Bleh. I can only hope that it'll be fixed soon; although I DO have the Cinelerra source code, I can't think of a single place to start to find out why YUVA 16-bit colormodels are broken. I suppose I could start in the rendering engine code, but chances are I wouldn't understand enough of it to do anything.
---
On the filter front:
Nada. Finals are eating up my time, so I've not made progress at all towards figuring out Cinelerra filter architectures, much less porting any of Donald Graft's uber-cool video-fixer-upper filters to Cinelerra. Which sucks, because without that, I can't really start the actual editing and compositing process.
Well, I _could_ use Cinelerra's built-in deinterlace filter, but it's on par with Premiere's deinterlacer (read: ick).
Speaking of finals: My Japanese final kicked ass. Woo. - yeah. 2002-11-18 00:37:36 Our Logic Design project kicked ass. That is all.
-
oops
2002-11-13 02:19:55
Correct previous journal entry:
"frame length" ---> "length of a clip on the timeline in real-time"
oops.
Also, 10th and final week here at Rose SUCK. Badly. My logic-design project is ALMOST DONE, except for those pesky few glitches that take damned never forever to debug. Oh yeah, it's due Thursday. ARGH.
Chemistry lab report: Been working on it for the past five hours. Have hit mental roadblock.
Seems that the only time that I write in this thing is when I've hit such roadblocks or when I'm feeling depressed. Whatever that says about me and my ego is up to you.
-
plugin authoring is fun
2002-11-10 15:25:38
Cinelerra 1.1 documents its plugin interface now, so I've been playing around with that recently. In about two hours I've figured out how to make a null transform realtime plugin, so I guess we're getting somewhere ;)
What I really hope to be able to do is port Donald Graft's Telecide and Smart Deinterlacer over. I'd love to be able to get Decimate in there as well, but that almost certainly isn't going to be a realtime plugin, since Cinelerra currently has the same limitation as VirtualDub when it comes to realtime plugin capabilities (i.e. cannot change frame length). Supposedly that's to be fixed in a later revision. We'll see.
Actually, come to think of it, Cinelerra can automatically decimate. The problem is that I don't know how well it works :P
This thing uses an assload of macros. Makes me shudder, but I suppose that there's a good reason for that. When you're looking for speed over size, you want to cut as many clock cycles as you possibly can, and shaving off those few clock cycles it takes to enter and exit a function can make a difference. Especially if you unroll loops.
That doesn't mean that it's not ugly, though.
- Cinelerra 1.1 is out 2002-11-08 17:25:37 Compiling now; let's see how this runs. Judging from what I've read in the program changelog, this release looks to kick some serious ass.
JOURNAL: trythil (💩 💩)
Current server time: Dec 21, 2025 10:55:20