holden
8th April 2004, 17:25
I have a few machines attached to my KVM which do little else than convert HD material. (On a rare occasion I use HDTVtoMPEG to tell me on what channels streams are located, but they're almost always on 0x21/24 in my capture configuration. Then...) I use dvd2avit3 (thanks to neuron2 and trbarry, I understand) to convert to d2v, Gordian Knot to tell me for certain the true size of the video and do cropping. (If the audio is not really worth a sumptuous media room, I'll convert it to VBRmp3 (GK@96-160kbps), else leave it as .ac3 surround.)
I keep up a window for a "work.avs" script and make very minor mods via Texturizer (fancy notepad) for each video. I also leave up a process running jonny's enc1.03, which is all set to process my work.avs via a few standard job lists which can quickly be brought to bear on the work.avs (lots less kb and mouse). (Also, output goes to separate hard drives so as to minimize disk thrashing.) I use a 2-pass video compression.
I usually use only mpeg2dec3, crop, trim, KernelDeint (many thanks, neuron2!!), decimate and LanczosResize (plus undot?). Surprisingly, KernelDeint often helps sharpen FILM.
I've got 1.8TB of video on my LAN (12 computers throughout the house). There's about 400 hours of HD, 500 hours of SD, 1000 hours of audio, and 10's of thousands of images. I play it all back in 1368x768 and encoded surround, as available, all quickly accessible. If the material that was originally MPEG2 had stayed in its original form, it would have taken up about 5TB. Thank goodness for MPEG4! (505, Xvid, 6mbps for HD, 1.5 for SD)
The results are all my personal video library, and I try my best not to do anything which would get me into trouble with the powers that be.
I don't say these things to brag. I'm laying it out there to get response and suggestions about anything anyone might see as to how I might do things better.
Anyway, here are my two main, current issues:
1. Film is a good thing--usually. The more recently film was shot, the fewer the problems. But in the conversion process for 1960-1990 vintage films, it seems to me some post-processing houses play some tricks which thwart well-intentioned re-encoding. Why, I'm not sure.
The results which are distressing were packaged as hybrid film-and-interlaced. I can work sections and make 'em OK, but it seems to me the .d2v generator should be able to work with AVISynth to let a script branch into two or three sections, one for interlaced, one for film, and possibly a third, progressive. I think I could handle it from there. Perhaps this hasn't been done because such detection is too unreliable(?). For hybrid stuff, it seems there are just too many back-and-forth sections to process them separately, manually. Perhaps there is AVISynth stuff to handle this of which I'm unaware and haven't seen discussed.
2. It seems to me there's a question of scale, especially for KernelDeint which may not be dealt with. Are motion compensation algorithms tuned for SD versus HD? Are they a percentage of the screen dimensions, or an absolute number of pixels? I still (though less often) get occasional "catches" (jerkiness) in my video when it seems to me the lens curvature or angle results in dramatically different speeds of motion in different portions of neighboring images. Are there built-in adjustments for HD versus SD? Otherwise, is there a way such a parameter (if needed, as I conjecture) can be tuned into these equations? Did I just miss this in the documentation?
Thanks in advance for your suggestions toward improvement.
HF
I keep up a window for a "work.avs" script and make very minor mods via Texturizer (fancy notepad) for each video. I also leave up a process running jonny's enc1.03, which is all set to process my work.avs via a few standard job lists which can quickly be brought to bear on the work.avs (lots less kb and mouse). (Also, output goes to separate hard drives so as to minimize disk thrashing.) I use a 2-pass video compression.
I usually use only mpeg2dec3, crop, trim, KernelDeint (many thanks, neuron2!!), decimate and LanczosResize (plus undot?). Surprisingly, KernelDeint often helps sharpen FILM.
I've got 1.8TB of video on my LAN (12 computers throughout the house). There's about 400 hours of HD, 500 hours of SD, 1000 hours of audio, and 10's of thousands of images. I play it all back in 1368x768 and encoded surround, as available, all quickly accessible. If the material that was originally MPEG2 had stayed in its original form, it would have taken up about 5TB. Thank goodness for MPEG4! (505, Xvid, 6mbps for HD, 1.5 for SD)
The results are all my personal video library, and I try my best not to do anything which would get me into trouble with the powers that be.
I don't say these things to brag. I'm laying it out there to get response and suggestions about anything anyone might see as to how I might do things better.
Anyway, here are my two main, current issues:
1. Film is a good thing--usually. The more recently film was shot, the fewer the problems. But in the conversion process for 1960-1990 vintage films, it seems to me some post-processing houses play some tricks which thwart well-intentioned re-encoding. Why, I'm not sure.
The results which are distressing were packaged as hybrid film-and-interlaced. I can work sections and make 'em OK, but it seems to me the .d2v generator should be able to work with AVISynth to let a script branch into two or three sections, one for interlaced, one for film, and possibly a third, progressive. I think I could handle it from there. Perhaps this hasn't been done because such detection is too unreliable(?). For hybrid stuff, it seems there are just too many back-and-forth sections to process them separately, manually. Perhaps there is AVISynth stuff to handle this of which I'm unaware and haven't seen discussed.
2. It seems to me there's a question of scale, especially for KernelDeint which may not be dealt with. Are motion compensation algorithms tuned for SD versus HD? Are they a percentage of the screen dimensions, or an absolute number of pixels? I still (though less often) get occasional "catches" (jerkiness) in my video when it seems to me the lens curvature or angle results in dramatically different speeds of motion in different portions of neighboring images. Are there built-in adjustments for HD versus SD? Otherwise, is there a way such a parameter (if needed, as I conjecture) can be tuned into these equations? Did I just miss this in the documentation?
Thanks in advance for your suggestions toward improvement.
HF