search

MilliVolt / matcher

GNU General Public License v3.0
1 stars 0 forks source link

thoughts on matching methods #25

Open myf opened 8 years ago

myf commented 8 years ago

our objective is to create good mismatches between two videos. The bridge we were thinking was to match video scene transition to audio beat time. Then we changed the methodology to two matching audio beat times.

However, it looks like that audio beat time is just a lot more finer than video scnen transiton. for an average 5 minute video there might be 20 or 30 scnene transitions but there are up to a few hundred audio beat times. In order to get a perfect match between these times are almost impossible if they are not using the same audio track.

I want to see if there's still some personally managable number of time signatures to work with (thinking every few second per time signature is a good respond time) and in he same time easier to compute

vr2262 commented 8 years ago

Our options for improving a2a matching:

  1. Reduce audio quality (may or may not have any impact)
  2. Mess with aubio parameters
  3. Mess with matcher parameters
  4. Reimplement matcher in Rust
vr2262 commented 8 years ago

Option 1 is easiest to try, followed by 2. I expect 3 to have a large impact. 4 is totally unknown but also orthogonal to the other options.

myf commented 8 years ago

@vr2262 if we can slow it down to a few sec per time sig that's good. im gonna play with tempo

myf commented 8 years ago

@vr2262 see my earlier comment, if we can do like 10 time signature per minutes i think it gives people enough time to respond to it

vr2262 commented 8 years ago

@myf trying to implement a "hopsize" was a failure... The score is very low unless the step happens to match the beat.