I came at this from the opposite direction I guess; for a long time I've been struggling with what to do with my off-air recordings from BBC HD (1080i h.264), which since they changed their encoder last summer have such crushing optimisation settings that a lot of players have problems playing them. This included the .mp4 "Native no-reencoding" export from EyeTV, which *nothing* could play. (QuickTime at a low framerate; VLC and XBMC lost sound, crashed HandBrake, for instance.)
For a long time I encoded them down to AppleTV (first using EyeTV's export, then HandBrake) accepting the huge drop in quality for playability through my AppleTV. But I kept the ones I cared most about around on a hard drive just in case I could do something more useful with them later. Now, with the most recent svn version, XBMC has resolved the various playback issues I was having (obvious interlacing-related combing artifacts, bad sound sync to name two) and is now my preferred player. It also, now, plays those raw recordings, with options like selecting the right (non audio-descriptive narration) audio track, disabling subtitles and so forth, but still has a few outstanding issues; a juddering start to playback, and problems with skipping forward or back. And the files are fairly large; reckon on about 6GB/hour. Nevertheless, just keeping and playing the raw eyetv program stream (or transport stream? i get confused) .mpg file is a workable option I'm still considering. Not having a slow encoding process to go through before being able to watch anything appeals.
So anyway after trying for a while to remux the original recordings into something that'll play better, I decided to reencode them all using HandBrake but to the same resolution. AppleTV won't play it, but XBMC will play happily, so will VLC, QuickTime/iTunes, Totem, actually even the PS3 which I could never get streaming from my vault before. And playback starts smoothly every time, and skipping back and forth works well. I decided to accept *some* quality reduction in exchange for compatibility; files that everything with enough oomph can actually play without problems. (I don't expect this to include the first-generation iPad when I get mine
) But of course I want to minimise the compromise on quality I have to make.
I used the Normal preset for this, just (manually for every file, grr) adding the AC3 Passthru track. The results look good, and if I'm sitting there trying hard to compare between them it's very difficult to find anything wrong; and oddly it's when I'm *not* trying so hard that I just find myself sometimes thinking, I can't put my finger on it but something about the image isn't quite *popping* my eyeballs as much as the originals. Subjective, unscientific, can't back it up, probably psychological (i know i'm watching the encoded-down version so I'm insisting there has to be something wrong with it).
But it meant having done the whole collection I was still wondering if I couldn't do it a better way. I kept wondering if I should be using High Profile rather than dowdy old Normal.
I'd gone with Normal because I just thought instinctively that the scaling inherent in High Profile (specifically in using Loose Anamorphic) *must* be reducing image quality. So I'd avoided it. But it always nagged at me that maybe I was missing something, in avoiding the setting that was so enticingly named "High Profile" and presumably for a reason.
I did some test encodes. comparing with the Normal encodes I'd already done - and to be honest I couldn't really see a difference. And I tried. I know, logically, there *must* be some loss through the High Profile scaling but it wasn't really perceptible compared to Normal, so the scaling algorithms must be very good.
I thought, well, maybe the way h.264 works, you don't really avoid the same sort of loss even if you keep the dimensions exactly the same; that perhaps the decode and re-encode together mean you still get the same sort of loss, so you might as well accept the scaling for the greater encoding efficiency of scaling for loose anamorphic, and the other high profile goodies. After all, the resulting file *is* slightly smaller.
I was about to post a new topic querying about this when I saw this thread, which seems to show the devs have come around logically to what I had been guessing instinctively at the start (avoid rescaling).
So it looks like sticking with Normal was the right thing to do. I think?
Some other observations... the original recordings are 1080i; ie: interlaced. Normal doesn't deinterlace in any way; no filters. High Profile decombs and detelecines. What do I see?
Playing back the Normal files; in QuickTime/VLC et al on my Mac (24", no scaling on playback) with my eyes inches away I can clearly see the 'combing' effect when the action on screen shows it up. But playback is perfect on the TV; the interlacing is just doing its thing there and everything is peachy. With High Profile, there's no 'combing' visible on the Mac, but on both the Mac and the TV, there's a barely perceptible loss of smoothness when things move. I have a test video, a studio-shot election programme, which includes a smooth camera pan across some dark vertical lines against a bright background. On High Profile, on TV and on the computer, this movement looks a bit staccato. With Normal, with interlacing allowed to do its business, it's smooth, but on the computer the combing is very evident.
So in sticking with Normal I'm making a decision to prefer best playback on the TV rather than the computer.
Finally I thought I'd try a 100% quality encode. Don't laugh. I know already.
I thought there's a chance this might be one of those rare times when it would be appropriate - as a way to essentially sanitise those original recordings, which are problematic for a lot of players, into ones which have no loss of quality but which play easily anywhere. The larger filesize, I thought, could be acceptable; eg: if it ended up making them comparable in size to the BBC HD recordings from before they changed to their current encoder last year.
Well, that didn't work out.
A ten-minute test recording produced a file nearly 7GB long, and neither Quicktime or XBMC could do anything with it. Unplayable. So that was the end of that. Worth a go so now I know.