Compressing HDTV for archival purposes

[jpiper]

I use MythTV for recording over cable ATSC/QAM HDTV. I'm running out of space on my drive, and I'd like to archive some recordings either to another HD or dual layer DVDs. The 6-7 GB/hr is too big for my tastes, but I don't want to sacrifice too much in terms of picture quality (eg, PAL DVD resolution is too low). I've started fooling around with ffmpeg to get the file sizes down.

For example:
ffmpeg -threads 2 -qscale 3 -vcodec mpeg4 -acodec copy -i 'input.mpg' 'output.avi'

I've been reading (on this and other VIDEOHelp forums/fora) that by using xViD (as in the ffmpeg call above), that qscale 3 should reduce file sizes by a factor of about 2 and qscale 4 should reduce sizes by a factor of 3 or so. That sounded great in theory, but the above call only reduced file sizes by a factor of 1.55 and qscale 4 by 1.85.

So then I read that using the x.264 codec should reduce the file sizes further for the same quality. First the qscale parameter doesn't seem to work, so I tried a bitrate of 7000k to see what quality that produces. I'm not sure of the results yet, since my machine only encodes this at 3 fps (as opposed to 37 using xViD).

This is my command for x.264 encoding (I was going to do 2-pass encoding):
ffmpeg -threads 2 -b 7000k -pass 1 -vcodec libx264 -acodec copy -i 'input.mpg' 'output.avi' -map 0:0 -map 0:1

So, I'm looking for some advise on what people are doing to compress their HDTV programs and whether my results seem reasonable. Also, if anyone can guide me on how to better use the x.264 compression for faster fps, I'd be appreciative. Although, my thinking as of now is with difference to xViD rather than x.264 unless we're talking about a 20% size reduction or more.

Thanks,
Jon

[redwudz]

jpiper,Do not post the same topic on several forums.
Please do not cross post, once is enough.

I deleted the other post as it was in the Mac ffmpegx forum.

I can move this post to our Linux forums if you like.
You may get more Linux specific answers there.

Moderator redwudz

[Soopafresh]

The fastest way would be splitting your mpeg files into 2 pieces, then saving each onto an individual DVD. You can do this with several different apps, such as ProjectX which works under Java.

The only way to speed up the transcoding speed of ffmpeg is to resize to a smaller dimension output file, ala

-s 1280x720


Note: Your first command line in your posting does not encode to XVID
Your h264 command line is assuming 2 pass, by the -pass 1 parameter. At 7000K bit rate, 1 pass should be fine, so no need to add the pass switch.



h264 encoding will take longer than Xvid, generally. You can still get excellent looking Xvid output from ffmpeg or mencoder.

[jpiper]

sorry about the double post...

I couldn't figure out from the documentation how to specify xvid explicitly and I read somewhere that it uses it by default when the vodec is specified as mpeg4. How would I use xvid and what does it use by default? [edit] I just figured it out: vcodec libxvid [/edit]. Looks like it only uses 1 processor for xvid encoding and it's only running at 4 fps with no resize... so we'll see what happens.

If I were to resize, should I also deinterlace?

I guess I shouldn't have said archiving, because that's not my exact goal - I'm really looking to keep these files on disk ideally (or many one one DVD), so I can play them on demand more or less.

Thanks,
Jon

[Soopafresh]

I'd think your mpeg sources are probably film, in which case you'd want to perform IVTC (inverse telecine) rather than deinterlace. I don't know the precise syntax in ffmpeg to do this, but this zip file might contain the necessary info:

http://osnews.com/img/vegas/ffmpeg-templates.zip

[jpiper]

Using xvid produced a file even larger than the default mpeg4 encoder (3%). qscale 4 starts to produce noticable artifcts and doesn't even quite reduce the file size by half.

Using the x.264 encoder wih one pass at 3500k (to reduce the file size about the same as qscale 4 mpeg4) with only one pass is producing awful results. I won't know until tonight what it will look like after the second pass, but even in frames where there is very little motion, i see plenty of artifact, and while encoding I see only high q-values (motly in the 20-30 range). I haven't tried the interlacing yet to keep any confounders out of the picture. I'll try that after the second pass.

I'm not very excited by what I'm seeing, but I'm hopeful I'm just missing something.

Does this jive with other people's experience or am I doing something wrong?

Thanks, Jon