Document background classes
Users of classifiers (DTW or kNN) want to know when there is no gesture that matches what has been trained. We need to document how the possible use of a "background" class, or rejecting bad matches by thresholding.
Activity
@francisco Does this make sense?
moved from RapidLib#88 (moved)
added Backlog label
added Website label
This makes a lot of sense to us here at IRCAM.
This has been a big problem for a while, with no ideal solution for now.
I think the "background class" solution is dependent on the use case (we sometimes use some examples labeled "silence" that play this role, but they are not generated, we record them by hand).
Rejecting bad matches by thresholding is something we could implement.
An estimated "confidence" output value could be nice to have as well in some cases (which could be seen as a kind of normalized distance on which we could define a normalized threshold to reject bad matches).added UCD label
mentioned in issue #22 (closed)
