On Sat, Feb 26, 2005 at 12:07:32AM -0500, Chris Pinkham wrote:
> > I would rather see the commflag process start on the recording, say
> > after 10 minutes, that way the longest you would have to wait for the
> > flagging to end is 10 minutes after the recording ends (instead of the
> > length of time that your recording is). I'm not sure what this would
> > affect, or if it's even doable.
>
> This is coming to CVS in the very near future. I'm trying to get
> commercial flagging so that it can be started almost right after the
> recording starts, so it will run about 30 seconds or so behind the
> recording. This will help get commercial flagging finished quicker
> (possibly right behind the recording ending), as well as helping cut
> down on disk thrashing since if it stays only 30 seconds behind, the
> recording data will still be in the filesystem cache hopefully and not
> require rereading.
If you're looking for ideas (and I don't stupidly suggest what you already
said you were doing like before) what would be really interesting for the
while-still-recording viewer would be a few heuristics that could make
this even better. They involve non-linear scanning.
a) Modify the player to report where it is play every few minutes.
Don't bother commercial scanning stuff the viewer has already
watched, at least not yet. Go back and do that later.
b) If you have identified a break with good confidence, skip
a safe interval in series and movies (but not sports) where it
is highly unlikely a 2nd break would occur. Most programs follow
a rough pattern on the minimum length of a program segment, except
near the start and end of the program. This is not an absolute
rule, of course, but can tell you where to scan first. You would
eventually scan the entire program and note all the things you are
looking for such as logo, scene transitions, fade-to-blacks, aspect
changes, sound saturation changes -- whatever you are looking for.
These sorts of tricks could generate a rough first pass, to be used if needed
immediately during watching, to be filled in later. Though I don't know how
adaptable your existing algorithms are to non-linear scan.
You can also be really assisted by the human being, if you get feedback
from the player. If the human being accepts a commercial skip (ie. plays
almost none of the presumptive break at 1x speed) you have a hard point
to start from.
To get more detailed, I might be viewing a program without flagging.
I hit 'End' as we enter a commercial break and it tells me not flagged -- but
I have just told you I really think there's the start of a commercial
break where I hit 'End.' Now you notice I did FF or 30 second skip multiple
times and where I start playing again is very near the end of that break.
This tells you several things. First of all, I've already watched to this
point, so no point in scanning before it right now. Secondly, a commercial
break ends near the point, so you can skip about 5 minutes ahead and start
your scanning. In fact, as long as you can scan not too much slower than
real time, this is enough -- all the fancy tricks I just described are
not strictly needed, because you have a 5 minute head start on where I
am watching, and can flag the rest until you catch up with the recording
point.
If flagging is much slower than real time (as can be the case for HDTV
while the CPU is being used to decode the HDTV) then the 5 minute head
start is still a lot of help, and once you find the next break, skipping
5 minutes after it could keep you ahead of the game on this live-watching
pass. One could also, in this situation, do a faster pass without the
slowest transition heuristics, and go back and do the real full-bore pass
later on.
Just some ideas, some complex, some easier. If the player can be convinced
to save the user's "watch stream" (When he hits End, changes from
1x watching to FF/skip watching) the basic technique is pretty simple,
and the End key a good way to trigger the start of the flagging past
that point.
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
