Computing best detect method via manually marked channels

Where all the user feedback and suggestions for improvements go
Post Reply
wtgreen
Posts: 31
Joined: Fri Jan 27, 2006 2:16 am

Computing best detect method via manually marked channels

Post by wtgreen »

I love comskip but wish there was a better way to tune it. It's default settings work great for certain channels but it's off some on other channels and programs. I've tried to tune it's detection settings in the past but it's very time consuming running comskip with a variety of methods and then comparing the results.

One feature I think could be really useful is a special tuning mode. It would accept a video or series of videos along with manually marked commercial file(s) and then determine which detection method or set of methods produces a result closest to the manually marked commercial file(s). The output of comskip would then be the "best match" INI file and perhaps some results statistics too.

I see several advantages to a feature like this - more than just better detection.

1) It doesn't really require the user to understand the nuances of the different detection methods. What comskip does isn't trivial by any means, and I know you field tons of questions on how best to tune it and how it should be configured for certain channels or certain countries. It's complex stuff, but manually tagging a file isn't too difficult. By us manually tagging some videos and having comskip tell us what methods will match those results best, your users better leverage the brain power you've put into comskip without having to bug you for assistance.

2) A user could more easily create INI files for different channels. I could run a set of videos through from one channel, save the INI that works best for this set of vids, and then the same for another channel and save that INI.

3) Tuning this way will likely be faster for the user, because while it does take some time to manually mark the commercials, at least when they're done they can run comskip in "tuning" mode and walk away knowing when comskip finishes, they'll have the best detection mode determined. No need to change the detection settings, run comskip, compare the results, change settings again, run comskip, repeat. A person can spend all day doing this because a single run of comskip can take so long and you have to change the INI file between each run. If it had this special "tuning" mode though the user does the work up-front and then let's comskip do it's thing. It doesn't matter too much how long it takes because the user's input is complete. We just have to come back later and see what comskip came up with.

4) Hopefully less support for you. Like I mentioned above, it takes advantage of all the smarts you've built into comskip without requiring the user to really understand all it's complexity and it's nuances. And when a user asks you for help in determining what tuning method is best for them, it gives you a quick and easy means to put the work back on them. You can just say run comskip -w, manually correct the tags, and then run it in tuning mode to see what will work best for you.

Anyway, it's just a thought. Thanks again for all your time and effort in making comskip.

Tim
erik
Site Admin
Posts: 3368
Joined: Sun Aug 21, 2005 3:49 pm

Re: Computing best detect method via manually marked channel

Post by erik »

I did try is approach and failed.
Then I added some of the automatic tuning options like
max_volume=0
and the information generated in the log file at the end to suggest some settings.
These parameters are often related to the actual audio/video signal in the broadcast.
But in actual video's where conflicting options where needed too chose from to get the best result I have never been able to do this automatic.
Post Reply