Mailing List Archive

commercial flagging questions
Two prefacing statements: Until last fall I used Windows/GB-PVR, which may
color my expectations, and I did search the list archives, so chide my
search capabilities, not my non-searchiness.

Specs: Kubuntu 7.10, myth .21, X2 3800, 2G ram, 6150

1. How long should I expect commflagging to take? I have "all methods"
enabled. Comskip (used by GB-PVR and others) consistently took about 4
minutes on a 1 hours show, same hardware (generally pegging 1 core), and
uses more detection methods. Detection was at least as good. Commflag took
about 30 minutes for a 1-hours show last night, though it seemed to run at
about 20-25% (guessing half of one core due to nice?). I saw at least one
post that indicated it should be a bit faster than that, but I don't know on
what hardware. hdparm shows good speed. If others are getting much better
times, then I'll look for something external to Myth. (I realize it can run
real-time, but for now I want to resolve this question.)

2. Is there any way to see the commercial breaks without editing the
recording and loading the breaks as a cutlist (Z)? (Did I characterize that
right?) Doing so seems to change the behavior, notably it seems to autoskip
even with autoskip off. My search for an OSD with commercial breaks turned
up "yeah, that'd be cool". Yup. Mostly I am interested in this because of
#3:

3. It looks like the only commercial-skipping keys are Z and Q, and they
always jump even if the position is nowhere near a commercial. My wife (and
thus I) had a problem the other day because she skipped at a commercial, and
it jumped 15 minutes or so. What happened was that first break was not
marked, so it jumped to the next break further into the show. I have found
somewhat of a solution in turning the auto-notify on with the blue osd.
However, now we have to watch for the popup, and if there's no commercial
use one key (right) to jump, and use another (Z) if there is a break. I
find mvpmc's skipping more natural: skip right is a 30-second skip, unless
the position is within a commercial break, and then the whole break is
skipped. With that one just has to know "jab this button until commercials
go away". Assuming the keyboard reference on the wiki is complete, I'm
guessing this is not possible? I also did not see any setup options.

Thanks,
John

* Comskip url: http://www.kaashoek.com/comskip
Code is available via SVN - don't know about the license. Could be
interesting to compare notes.
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16311880.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
bungle wrote:
> Two prefacing statements: Until last fall I used Windows/GB-PVR, which may
> color my expectations, and I did search the list archives, so chide my
> search capabilities, not my non-searchiness.
>
> Specs: Kubuntu 7.10, myth .21, X2 3800, 2G ram, 6150
>
> 1. How long should I expect commflagging to take? I have "all methods"
> enabled. Comskip (used by GB-PVR and others) consistently took about 4
> minutes on a 1 hours show, same hardware (generally pegging 1 core), and
> uses more detection methods. Detection was at least as good. Commflag took
> about 30 minutes for a 1-hours show last night, though it seemed to run at
> about 20-25% (guessing half of one core due to nice?). I saw at least one
> post that indicated it should be a bit faster than that, but I don't know on
> what hardware. hdparm shows good speed. If others are getting much better
> times, then I'll look for something external to Myth. (I realize it can run
> real-time, but for now I want to resolve this question.)
>
Allow your jobs to use more processor if you want it to use more
processor. There's an option in mtyhtv-setup. I believe the default is
"Low" which is what you're seeing. You can choose to peg your processor
if you like. :) Commflagging generally happens realtime for me on an
AMD64 X2 4200+. About a minute after the program finishes, the commflag
job is done.
> 2. Is there any way to see the commercial breaks without editing the
> recording and loading the breaks as a cutlist (Z)? (Did I characterize that
> right?) Doing so seems to change the behavior, notably it seems to autoskip
> even with autoskip off. My search for an OSD with commercial breaks turned
> up "yeah, that'd be cool". Yup. Mostly I am interested in this because of
> #3:
>
Coorect. Not implemented to show commflags in the OSD. You're right - it
would be cool. :)
> 3. It looks like the only commercial-skipping keys are Z and Q, and they
> always jump even if the position is nowhere near a commercial. My wife (and
> thus I) had a problem the other day because she skipped at a commercial, and
> it jumped 15 minutes or so. What happened was that first break was not
> marked, so it jumped to the next break further into the show. I have found
> somewhat of a solution in turning the auto-notify on with the blue osd.
> However, now we have to watch for the popup, and if there's no commercial
> use one key (right) to jump, and use another (Z) if there is a break. I
> find mvpmc's skipping more natural: skip right is a 30-second skip, unless
> the position is within a commercial break, and then the whole break is
> skipped. With that one just has to know "jab this button until commercials
> go away". Assuming the keyboard reference on the wiki is complete, I'm
> guessing this is not possible? I also did not see any setup options.
>
I generally use a combination of 4 keys to skip through programs.
Advance goes ahead 30 seconds - no matter what. Replay goes back 5
seconds - no matter what. Skip+ goes to the next commflag point (either
beginning of the next or end of the current), Skip- goes to the previous
commflag point (either end of the last or beginning of the current).

I also use auto-skip so I generally never touch the remote when watching
a program and I don't see any commercials at all. If it skips ahead too
much (as it does sometimes), I grab the remote, hit Skip- and it will go
back to where the big bad skip happened and start playing video
normally. If there is a commercial in there somewhere, I Skip+ and it
goes to the end and I continue watching.

My 10 year old doesn't really get this concept, so if something skips
too much or not enough, she just uses the Advance and Replay keys to go
forward and back as needed until she's at the right place.

-Brad

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
> 1. How long should I expect commflagging to take? I have "all methods"
> enabled. Comskip (used by GB-PVR and others) consistently took about 4
> minutes on a 1 hours show, same hardware (generally pegging 1 core), and
> uses more detection methods. Detection was at least as good. Commflag
took
> about 30 minutes for a 1-hours show last night, though it seemed to run at
> about 20-25% (guessing half of one core due to nice?). I saw at least one
> post that indicated it should be a bit faster than that, but I don't know
on
> what hardware. hdparm shows good speed. If others are getting much
better
> times, then I'll look for something external to Myth. (I realize it can
run
> real-time, but for now I want to resolve this question.)
>

Wow, i didnt realize GB was that fast, i wonder how the other PVR software
rate in speed/accuracy.


> > recording and loading the breaks as a cutlist (Z)? (Did I characterize
> that
> > right?) Doing so seems to change the behavior, notably it seems to
> autoskip
> > even with autoskip off. My search for an OSD with commercial breaks
> turned
> > up "yeah, that'd be cool". Yup. Mostly I am interested in this because
> of
> > #3:
> >
> > 2. Is there any way to see the commercial breaks without editing the
> Coorect. Not implemented to show commflags in the OSD. You're right - it
> would be cool. :)
>


I have always thought the OSD and entire Myth interface leaves a lot to be
desired. It pretty feature-less but it works. I wish the OSD had a view
like TiVo (Where if timeshifting you see what portion of the program you
have). But there are limited devs and if I had to choose between a nicer
OSD/"Prettyness" or a useful feature (ie Storage Groups or Input Groups) i
would choose the later.

Mitchell
Re: commercial flagging questions [ In reply to ]
On Wed, Mar 26, 2008 at 2:43 PM, bungle <nabble.e.wuuza@xoxy.net> wrote:
> 2. Is there any way to see the commercial breaks without editing the
> recording and loading the breaks as a cutlist (Z)? (Did I characterize that
> right?) Doing so seems to change the behavior, notably it seems to autoskip
> even with autoskip off. My search for an OSD with commercial breaks turned
> up "yeah, that'd be cool". Yup. Mostly I am interested in this because of
> #3:

I'm pretty sure you can also delete all the cut points with key, but
I'm not sure which. So you could load the cut points to look and then
delete them again.

> 3. It looks like the only commercial-skipping keys are Z and Q, and they
> always jump even if the position is nowhere near a commercial. My wife (and
> thus I) had a problem the other day because she skipped at a commercial, and
> it jumped 15 minutes or so. What happened was that first break was not
> marked, so it jumped to the next break further into the show. I have found
> somewhat of a solution in turning the auto-notify on with the blue osd.
> However, now we have to watch for the popup, and if there's no commercial
> use one key (right) to jump, and use another (Z) if there is a break. I
> find mvpmc's skipping more natural: skip right is a 30-second skip, unless
> the position is within a commercial break, and then the whole break is
> skipped. With that one just has to know "jab this button until commercials
> go away". Assuming the keyboard reference on the wiki is complete, I'm
> guessing this is not possible? I also did not see any setup options.

There's a comm-skip limit that you can set so you won't skip over long
commercials (sorry I don't remember which menu it's in). I have mine
set to 5 minutes. It gives you a warning if the commercial is longer
than that and will skip if you hit the skip button again. Also if you
hit the comm-skip back button (Q on the keyboard) within a few seconds
of skipping a commercial you'll skip back to the point you skipped
from instead of the previous cut point.

>
> Thanks,
> John
>
> * Comskip url: http://www.kaashoek.com/comskip
> Code is available via SVN - don't know about the license. Could be
> interesting to compare notes.
> --
> View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16311880.html
> Sent from the mythtv-users mailing list archive at Nabble.com.
>
> _______________________________________________
> mythtv-users mailing list
> mythtv-users@mythtv.org
> http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
>



--
It was supposed to be so Easy.
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Brad DerManouelian-2 wrote:
>
> Allow your jobs to use more processor if you want it to use more
> processor. There's an option in mtyhtv-setup. I believe the default is
> "Low" which is what you're seeing. You can choose to peg your processor
> if you like. :) Commflagging generally happens realtime for me on an
> AMD64 X2 4200+. About a minute after the program finishes, the commflag
> job is done.
>

Ah, I did not see this (there are a lot of menus, but options are good). I
will see how much difference it makes (part of the reason I got a
dual-core). I assume you mean you run commflagging while still recording?
I figured eventually I would try that, but I wanted to make sure I didn't
have some system problem this was revealing.


Brad DerManouelian-2 wrote:
>
> Coorect. Not implemented to show commflags in the OSD. You're right - it
> would be cool. :)
>

Standard OSD in GB-PVR had this - it was pretty nice. I'm sure it's a
simple as borrowing the code from the cutlist OSD ;) Seriously, I can live
without. It was a good graphical representation of how much of my life I
was taking back, though.



Fred Squires wrote:
>
> There's a comm-skip limit that you can set so you won't skip over long
> commercials (sorry I don't remember which menu it's in). I have mine
> set to 5 minutes. It gives you a warning if the commercial is longer
> than that and will skip if you hit the skip button again. Also if you
> hit the comm-skip back button (Q on the keyboard) within a few seconds
> of skipping a commercial you'll skip back to the point you skipped
> from instead of the previous cut point.
>


I saw that, but wasn't sure if it referred to not flagging commercial breaks
that long or what. The other is a good tip, also.

Thanks
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16314754.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
bungle wrote:
> Ah, I did not see this (there are a lot of menus, but options are good). I
> will see how much difference it makes (part of the reason I got a
> dual-core). I assume you mean you run commflagging while still recording?
> I figured eventually I would try that, but I wanted to make sure I didn't
> have some system problem this was revealing.
>
Yes, I run commercial flagging while programs are recording. (another
setting!!) It's great when you know an hour long program starts at 8,
you can sit down at 8:20 or so and watch the whole thing while
commercials automatically skip by. By the time 9:00 rolls around, you're
done watching and saved yourself 20 minutes before the program started
to do other stuff - like watch a half hour program that was already
recorded. ;)

If you want to get an idea of what kind of speed you're getting, check
the Mythweb Status page while commercial flagging is running. In the Job
Queue section, hover over the job that's running and you'll see the
progress including FPS on that video. Right now I'm getting 54fps on a
SD recording. I think I get about 15-20fps on HD content. Oh, and top
says I'm using 12-15% of one core so I must have it set pretty low.
> Brad DerManouelian-2 wrote:
>
>> Coorect. Not implemented to show commflags in the OSD. You're right - it
>> would be cool. :)
>>
>>
>
> Standard OSD in GB-PVR had this - it was pretty nice. I'm sure it's a
> simple as borrowing the code from the cutlist OSD ;) Seriously, I can live
> without. It was a good graphical representation of how much of my life I
> was taking back, though.
>
It gets a little wonky when you're flagging while watching. You have to
keep querying the flagpoints every time the OSD is displayed and display
them in relation to the end of the program. Unlike other OSD's Myth's
progress bar is relative to the end of the video, not to the end of the
recording schedule. So you don't see the full length of the video
represented on the progress bar, just the video that's been recorded.
Nice for when the actual video doesn't match the schedule or you have a
partial video that you've edited down (i.e. cut commercials) but not so
nice for displaying commflag points accurately along the progress bar.
Not impossible, just never been done in myth before.

-Brad

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Brad DerManouelian-2 wrote:
>
> If you want to get an idea of what kind of speed you're getting, check
> the Mythweb Status page while commercial flagging is running. In the Job
> Queue section, hover over the job that's running and you'll see the
> progress including FPS on that video. Right now I'm getting 54fps on a
> SD recording. I think I get about 15-20fps on HD content. Oh, and top
>

I actually did see that it reported that - pretty cool. I was getting 70+
on SD - sounds like everything is healthy. I'll jack up the priority and
see what happens.

John
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16315944.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Mitch Gore wrote:
>
>> enabled. Comskip (used by GB-PVR and others) consistently took about 4
>> minutes on a 1 hours show, same hardware (generally pegging 1 core), and
> Wow, i didnt realize GB was that fast, i wonder how the other PVR software
> rate in speed/accuracy.
>

I think BTV and Sage both use comskip, too, so I'm not sure there are many
others. Comskip does have some cool GUI tools. The detection methods
available in the config (along with a LOT of other options) are: 1=black
frame, 2=logo, 4=scene change, 8=fuzzy logic, 16=closed captions, 32=aspect
ratio, 64=silence, 128=cutscenes, though mine (and the default) was 111.


Mitch Gore wrote:
>
> I have always thought the OSD and entire Myth interface leaves a lot to be
> desired. It pretty feature-less but it works. I wish the OSD had a view
>

GB-PVR is probably more geeky than MCE or Sage, but not like Myth (in the
substance over style, tweaky way). Myth has some pretty nice themes, but I
can see a point about the organization. (Going more OT) Myth has so many
features, and it's organized pretty logically, but in such a way that things
get fairly deep in menus. GB-PVR is a bit more simplified so things seem
easier to find. I like using both (albeit fairly simply - record, watch,
delete; remote music via MVP), and they both have their strengths, but
GB-PVR was not open source, which was one factor, but also only ran on
Windows. Hardware supporting W2k was harder to find, and I did not like the
upgrade options (among other factors).


--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16319744.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
* On Wed Mar 26, 2008 at 05:25:12PM -0700, bungle wrote:
> Mitch Gore wrote:
> > Wow, i didnt realize GB was that fast, i wonder how the other PVR software
> > rate in speed/accuracy.

> I think BTV and Sage both use comskip, too, so I'm not sure there are many
> others.

Comskip was also originally a port of my commercial flagging code in
Myth. At one point, I even had a Linux version of it compiled and
running using a Makefile I built. Since then a lot or most of it has
been rewritten, but it all started with MythTV. I know they looked at
using non-open source codecs, so that could be one reason comskip is
faster. I also think it skips frames in order to run faster. The
current flagging code in Myth has quite a few things in it to speed up
detection on a per-frame basis, but it does look at every frame. It
also is single-threaded. There was a project during the 2006 SoC to
add multithreaded flagging, but it never materialized into anything.

> available in the config (along with a LOT of other options) are: 1=black
> frame, 2=logo, 4=scene change, 8=fuzzy logic, 16=closed captions, 32=aspect
> ratio, 64=silence, 128=cutscenes, though mine (and the default) was 111.

So the extras they give you are fuzzy logic (??), closed captions,
silence, and cut scenes. Since people keep suggesting things to work on
for the flagger in Myth, I'll say that given the current code design and
where the most bang for the buck is, I'd recommend looking at the closed
captions and the silence detection. Just thought I'd throw that in
since this thread may invariably turn into how to improve Myth's
flagger.

--
Chris
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
I pumped the priority to high and commflag took about 7-8 minutes for a one
hour show, so that's pretty good. I guess it was a lot more "nice" before
than I thought. I thought 25% was 1/2 of one core, but I guess it's 1/4
since now it ran at close to 100% on top. One core full-bore must be 100%
on Linux (200 total), whereas on Windows each is 50% (100 total).
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16320267.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On Wed, Mar 26, 2008 at 8:56 PM, Chris Pinkham <cpinkham@bc2va.org> wrote:
> Since people keep suggesting things to work on
> for the flagger in Myth, I'll say that given the current code design and
> where the most bang for the buck is, I'd recommend looking at the closed
> captions and the silence detection. Just thought I'd throw that in
> since this thread may invariably turn into how to improve Myth's
> flagger.

I've actually experimented with trying to use close captions to help
commercial flagging, though i did it with the experimental flagger,
not the classical one. It was more a proof of concept thing so to
start with i used ccextractor to extract the closed captions and then
attempted to match the closed captions up to frames based on the
timestamps. Then i tried to apply the same Bayesian type filtering
used in spam detection. It worked horribly, so i tried different ways
of tokenizing the captions. Still horrible. Now there are many things
i could have been doing wrong, possibly not getting the captions lined
up to the correct frames, or not tokenizing well, but as i looked over
my data it seemed that there were no words that were good indicators
of commercials. Now it is possible someone else could use them to
help, but i though i would share my experience.
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On 03/26/2008 02:43 PM, bungle wrote:
> 3. It looks like the only commercial-skipping keys are Z and Q, and they
> always jump even if the position is nowhere near a commercial. My wife (and
> thus I) had a problem the other day because she skipped at a commercial, and
> it jumped 15 minutes or so.

Hit the button to skip to next commercial break marker (End or Z).
Notice that Myth says, "Skipped 15:32." Hit the button to skip to
previous commercial break marker (Home or Q). No matter where you were
when you hit the Z key, if you hit Q within a few seconds (3? I think),
you skip back to where you were when you hit Z. Therefore, at this
point, you know where you're at (if you hit the skip button when the
commercial break started, you can now use the appropriate skip
(Left/Right)/jump (Up/Down) keys to skip the break manually).

By adjusting your skip/jump amounts in your playback group configuration
appropriately, you can manually skip a commercial break very easily (and
in very few button presses).

Of course, you could use FFwd/Rew, but, IMHO, they're a waste when you
have skip/jump.

Mike

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On 03/26/2008 11:35 PM, Paul wrote:
> On Wed, Mar 26, 2008 at 8:56 PM, Chris Pinkham <cpinkham@bc2va.org> wrote:
>
>> Since people keep suggesting things to work on
>> for the flagger in Myth, I'll say that given the current code design and
>> where the most bang for the buck is, I'd recommend looking at the closed
>> captions and the silence detection. Just thought I'd throw that in
>> since this thread may invariably turn into how to improve Myth's
>> flagger.
>>
> I've actually experimented with trying to use close captions to help
> commercial flagging, though i did it with the experimental flagger,
> not the classical one. It was more a proof of concept thing so to
> start with i used ccextractor to extract the closed captions and then
> attempted to match the closed captions up to frames based on the
> timestamps. Then i tried to apply the same Bayesian type filtering
> used in spam detection. It worked horribly, so i tried different ways
> of tokenizing the captions. Still horrible. Now there are many things
> i could have been doing wrong, possibly not getting the captions lined
> up to the correct frames, or not tokenizing well, but as i looked over
> my data it seemed that there were no words that were good indicators
> of commercials. Now it is possible someone else could use them to
> help, but i though i would share my experience.

Which, I think all boils down to: CC doesn't work so well (at least, in
the US) because anymore, many (most?) commercials now have captions,
too. I'm guessing the implementation in comskip was done at a time when
assuming the show had CC and the commercials didn't was a valid
assumption, so it uses that simplistic approach.

Mike

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
* On Wed Mar 26, 2008 at 07:20:36PM -0700, bungle wrote:
> I pumped the priority to high and commflag took about 7-8 minutes for a one
> hour show, so that's pretty good. I guess it was a lot more "nice" before
> than I thought. I thought 25% was 1/2 of one core, but I guess it's 1/4
> since now it ran at close to 100% on top. One core full-bore must be 100%
> on Linux (200 total), whereas on Windows each is 50% (100 total).

The 'Low' CPU setting also sleeps a little every frame so it's more than
just niced. The 'Medium' setting is just niced and 'High' is
full-bore (no sleep and no nice). The ~realtime flagging alters its
speed as it runs in order to stay close behind real time while not
catching up too close.

--
Chris
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
* On Thu Mar 27, 2008 at 12:09:41AM -0400, Michael T. Dean wrote:
> Which, I think all boils down to: CC doesn't work so well (at least, in
> the US) because anymore, many (most?) commercials now have captions,
> too. I'm guessing the implementation in comskip was done at a time when
> assuming the show had CC and the commercials didn't was a valid
> assumption, so it uses that simplistic approach.

I believe that comskip has a list of phrases that it uses to weight
commercials versus non-commercials. So it's something that gets better
the more you train it with these phrases. Sort of like what people
are talking about doing with the video in a commercial, but using
the much easier method of just looking at the CC data. If the phrase
"Milk, it does the body good" pops up, then give that block a few
points extra because it's probably a commercial, but if "book 'em
Danno!" is heard then it's probably not.

--
Chris
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On Wed, Mar 26, 2008 at 11:59 PM, Michael T. Dean <mtdean@thirdcontact.com>
wrote:

> On 03/26/2008 02:43 PM, bungle wrote:
> > 3. It looks like the only commercial-skipping keys are Z and Q, and they
> > always jump even if the position is nowhere near a commercial. My wife
> (and
> > thus I) had a problem the other day because she skipped at a commercial,
> and
> > it jumped 15 minutes or so.
>
> Hit the button to skip to next commercial break marker (End or Z).
> Notice that Myth says, "Skipped 15:32." Hit the button to skip to
> previous commercial break marker (Home or Q). No matter where you were
> when you hit the Z key, if you hit Q within a few seconds (3? I think),
> you skip back to where you were when you hit Z. Therefore, at this
> point, you know where you're at (if you hit the skip button when the
> commercial break started, you can now use the appropriate skip
> (Left/Right)/jump (Up/Down) keys to skip the break manually).
>

Additionally you can set the maximum number of minutes that it will skip.
Set it to 5 and it pretty much eliminates all extra long jumps like that.
Re: commercial flagging questions [ In reply to ]
On 03/27/2008 05:58 AM, Chris Pinkham wrote:
> * On Thu Mar 27, 2008 at 12:09:41AM -0400, Michael T. Dean wrote:
>
>> Which, I think all boils down to: CC doesn't work so well (at least, in
>> the US) because anymore, many (most?) commercials now have captions,
>> too. I'm guessing the implementation in comskip was done at a time when
>> assuming the show had CC and the commercials didn't was a valid
>> assumption, so it uses that simplistic approach.
>>
> I believe that comskip has a list of phrases that it uses to weight
> commercials versus non-commercials. So it's something that gets better
> the more you train it with these phrases. Sort of like what people
> are talking about doing with the video in a commercial, but using
> the much easier method of just looking at the CC data. If the phrase
> "Milk, it does the body good" pops up, then give that block a few
> points extra because it's probably a commercial, but if "book 'em
> Danno!" is heard then it's probably not.

Cool. That's pretty impressive (though I think I'd still prefer to see
work on the audio-based detection schemes).

Mike
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Michael T. Dean wrote:
>
> too. I'm guessing the implementation in comskip was done at a time when
> assuming the show had CC and the commercials didn't was a valid
> assumption, so it uses that simplistic approach.
>

I'm not sure. My (now older) version had a comskip.dictionary file that
appeared to be in two sections - one with some words you'd expect to find in
shows, and one that had ad-type words, like zocor, geico, nexium, etc. So I
think it must have looked for some threshold of words. I don't see that
file in the current download, so I don't know what it does now.
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16325239.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Chris Pinkham wrote:
>
> using non-open source codecs, so that could be one reason comskip is
> faster. I also think it skips frames in order to run faster. The
>

According to my latest results, it seems less than twice as fast, so
skipping every other frame (which probably wouldn't miss much) could easily
account for a lot of that. (Before I knew how to control the job queue I
wasn't really comparing the same benchmarks.) I've only compared SD, not
HD.


Chris Pinkham wrote:
>
> captions and the silence detection. Just thought I'd throw that in
> since this thread may invariably turn into how to improve Myth's
> flagger.
>

I haven't had nearly as much experience with commflag, but it seems to do a
very good job. We watch a lot of shows like Mythbusters that have a little
segue bit or whatever it is called, and both do nearly perfectly with those.
When comskip missed something, it would usually be a spot right at the
beginning or ending of a break, or sometimes one or two in the middle. I've
had a few cases where commflag missed the mark and it ignored the whole
break, but I'd say most of the time it gets things right.

(I'm not necessarily advocating any of this stuff, just passing along the
info.)
Comskip did have more ways to tune scanning to your signal (based on my
current knowledge of Myth), such as running the GUI or looking at the logs
to see what kind of numbers it generates, and then tuning the ini file with
settings like "max_brightness" and "min_silence". (Digital STB would
presumably have blacker black than analog cable.) It also had settings to
control whether the last little bit of preview or credits after the last
break is show or commercial (which is really a per-show thing anyway).
Frankly, though, I didn't use any of that much, as it was interesting but
not something I wanted to spend all day on. It doesn't take long to skip a
missed break, and if I wanted to save the show it's pretty easy to edit the
breaks. The defaults worked well enough usually.

John

Aforementioned settings: http://www.kaashoek.com/files/comskip.ini
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16325252.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Chris Pinkham wrote:
>
> using non-open source codecs, so that could be one reason comskip is
> faster. I also think it skips frames in order to run faster. The
>

According to my latest results, it seems less than twice as fast, so
skipping every other frame (which probably wouldn't miss much) could easily
account for a lot of that. (Before I knew how to control the job queue I
wasn't really comparing the same benchmarks.) I've only compared SD, not
HD.


Chris Pinkham wrote:
>
> captions and the silence detection. Just thought I'd throw that in
> since this thread may invariably turn into how to improve Myth's
> flagger.
>

I haven't had nearly as much experience with commflag, but it seems to do a
very good job. We watch a lot of shows like Mythbusters that have a little
segue bit or whatever it is called, and both do nearly perfectly with those.
When comskip missed something, it would usually be a spot right at the
beginning or ending of a break, or sometimes one or two in the middle. I've
had a few cases where commflag missed the mark and it ignored the whole
break, but I'd say most of the time it gets things right.

(I'm not necessarily advocating any of this stuff, just passing along the
info.)
Comskip did have more ways to tune scanning to your signal (based on my
current knowledge of Myth), such as running the GUI or looking at the logs
to see what kind of numbers it generates, and then tuning the ini file with
settings like "max_brightness" and "min_silence". (Digital STB would
presumably have blacker black than analog cable.) It also had settings to
control whether the last little bit of preview or credits after the last
break is show or commercial (which is really a per-show thing anyway).
Frankly, though, I didn't use any of that much, as it was interesting but
not something I wanted to spend all day on. It doesn't take long to skip a
missed break, and if I wanted to save the show it's pretty easy to edit the
breaks. The defaults worked well enough usually.

John

Aforementioned settings: http://www.kaashoek.com/files/comskip.ini

Sorry if this is a repost - I use Nabble and the list wasn't taking it for
some reason. (Perhaps my prior response about close captioning that had
some drug names in it describing what comskip looks for.)
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16325252.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
bungle <nabble.e.wuuza@xoxy.net> says:
> According to my latest results, [comskip] seems less than twice as
> fast, so skipping every other frame (which probably wouldn't miss
> much) could easily account for a lot of that.

The skip-every-other-frame idea does sound interesting; in this
non-coder's mind it sounds like an easy way to significantly improve
performance on a task that doesn't seem to require that every single
frame be examined. Heck, even one frame per second or half second
sounds usable. Or am I missing something obvious? It's not like
mythcommflag's performance is really lacking anything as things stand,
at least on my setup.

--
Frontend: P4 3.0GHz, 1.5TB software RAID 5 array
Backend: Quad-core Xeon 1.6GHz, 6.6TB sw RAID 6
Video inputs: Four high-definition over FireWire/OTA
Accessories: 47" 1080p LCD, 5.1 digital, and MX-600
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On Thu, Mar 27, 2008 at 11:05:54AM -0700, Yeechang Lee wrote:
> bungle <nabble.e.wuuza@xoxy.net> says:
> > According to my latest results, [comskip] seems less than twice as
> > fast, so skipping every other frame (which probably wouldn't miss
> > much) could easily account for a lot of that.
>
> The skip-every-other-frame idea does sound interesting; in this
> non-coder's mind it sounds like an easy way to significantly improve
> performance on a task that doesn't seem to require that every single
> frame be examined. Heck, even one frame per second or half second
> sounds usable. Or am I missing something obvious? It's not like
> mythcommflag's performance is really lacking anything as things stand,
> at least on my setup.

Indeed, is it reasonable to examine every Nth frame, and if it
looks like a transition has occurred, move back N frames and
examine every one for the break?

-dsr-

--
Restore our Constitutional rights.

http://tao.merseine.nu/~dsr/eula.html is hereby incorporated by reference.
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
On Thu, Mar 27, 2008 at 2:49 PM, Dan Ritter <dsr-myth@tao.merseine.nu> wrote:
> On Thu, Mar 27, 2008 at 11:05:54AM -0700, Yeechang Lee wrote:
> > bungle <nabble.e.wuuza@xoxy.net> says:
> > > According to my latest results, [comskip] seems less than twice as
> > > fast, so skipping every other frame (which probably wouldn't miss
> > > much) could easily account for a lot of that.
> >
> > The skip-every-other-frame idea does sound interesting; in this
> > non-coder's mind it sounds like an easy way to significantly improve
> > performance on a task that doesn't seem to require that every single
> > frame be examined. Heck, even one frame per second or half second
> > sounds usable. Or am I missing something obvious? It's not like
> > mythcommflag's performance is really lacking anything as things stand,
> > at least on my setup.
>
> Indeed, is it reasonable to examine every Nth frame, and if it
> looks like a transition has occurred, move back N frames and
> examine every one for the break?

That might work. I don't think only examining once per second or half
second would be a good choice.
Usually if I manually cut commercials skipping by seconds will find a
blank frame, and if not half seconds will, but sometimes the blank
frames at the beginnings and ends of commercial breaks are only a few
frames long or even as little as a single frame. That might not mess
up all the detection methods but I don't think blank frame detection
would like it very much.

--
It was supposed to be so Easy.
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: commercial flagging questions [ In reply to ]
Yeechang Lee wrote:
>
> The skip-every-other-frame idea does sound interesting; in this
> non-coder's mind it sounds like an easy way to significantly improve
> performance on a task that doesn't seem to require that every single
> frame be examined. Heck, even one frame per second or half second
> sounds usable. Or am I missing something obvious? It's not like
>

I would suspect the biggest problem would be black-frame detection, since
there probably aren't very many of those frames between segments, and you
wouldn't want to skip them. Not sure, though.
--
View this message in context: http://www.nabble.com/commercial-flagging-questions-tp16311880s15552p16337192.html
Sent from the mythtv-users mailing list archive at Nabble.com.

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users