On Sat, 2008-01-12 at 16:09 +0100, Toni Pongratz wrote: > Hi all, Hi,
> First things first: I'm using transcode 1.0.4, and these problems occured > with > both the x86-64 packman openSUSE 10.3 build as well as with a self-compiled > version. I also tried 1.0.5, with no difference. Please provide the full command line used, possibily in the various stages tried, in order to better understand the transcode (wrong) behaviour. > I have here a lot of raw interlaced video streams in packed YUY2 format that > I > need to encode and render both some text and a transparent png image into it. > _In theory_, transcode could handle all that in just one run. In practice, it > produces garbage output. Are these streams in DV format? Or maybe uncompressed YUY2 frames packed in AVI or anything else? > The first thing is: the yuy2toyv12 filter (which is actually misspelled as > yuy2tov12 in the manpage) Fixed > produces garbage video when added. If needed, I could provide sample pictures > of the output. For a better diagnosis of this issue I need the commandline used, as outlined above. > Because of that, I used mencoder with a -vf scale,format=yv12, which runs > smoothly and produces correct output (at this point, I'm also cropping the > video to remove black borders. I believe transcode can do that, too, but > mencoders crop filter is simpler to understand ;) ). Uhm, I should really improve detectclipping filter and its documentation as well :) > So now I had video that transcode can cope with and all would be fine. But > no, > the next problem is the logo filter. It will _not_ handle alphatransparent > png images correctly. With the original logo that I used it simply rendered > only a fraction of the image. After I've tried changing the logo, and ended > up refining it, it won't be blended into the video at all. No trace of it. > GIF works, but is absolutely not an option. filter_logo was seriously improved during 1.1.0 cycle, maybe you could give a try using transcode 1.1.0-alpha6. > So now it gets really ugly. I had to export it frame-per-frame, then > composite > the logo onto every image using imagemagick (which is all done by a script, > of course), and then import it again and re-create the video. Since this > would result in an upsidedown, red-blue-swapped video, I added -z -k. Those anomalies are supposed to be fixed as well in 1.1.0 > Except, now the -k switch produces garbage. And not only that, importing the > audiostream will not really work either. transcode will speed up the sound > (the video is using 30fps), probably on the ground that it's for 25fps video > or something. I've fiddled around with some fps switches, got no real > results. That's hard maybe seamples will be needed to fix that, and this raise various problems, first and most important being copyright issues that I absolutely want to avoid. > Which means I have to merge the audio stream into the video using avimerge > (_yet_ another step). > To correct the colors I then went on to to use mencoder with -vf > rgb2bgr=swap,scale,format=yv12 (to get rid of that whole RGB-stuff altogether > again), and then finally encoded the video to xvid. The result is all fine, > and I could probably cut one step by using mencoder to encode it, but still - > this script takes about 80 minutes to produce acceptable results for, if > you'll excuse me, a friggin' 2 minute video. No, this is definitively not acceptable. Let's see how this can be improved. Bests, -- Francesco Romani // Ikitt [ Out of memory. ~ We wish to hold the whole sky, ~ But we never will. ]
