"west side story" is labeled by me, and some the features are extracted. we will train some models built by them. Also, sutony has trained 2 models by "rent" features with 1000 iterations using multiboost toolkit, the outcome is as below:
Error
class 1: 43.5%
class 2: 57%
class 3: 41.38%
class 4: 58.48%
overall: 50.09%
Error
class 1: 32.82%
class 2: 46.8%
class 3: 33.64%
class 4: 47.08%
overall: 40.09%
the result seems strange. the latter model is always better than the previous one in all classes. we will try to figure out what happen.
multiboost toolkit:
http://www.iro.umontreal.ca/~casagran/multiboost.html#
2 則留言:
We didn't use all of the haar-like features to train the model, because we were afraid that training procedure might take to much time. However, it seems that the training time is acceptable, so maybe I'll try to use more haar-like features to see if the performance could be improved.
Although the error rate seems so bad (about 50% for each class), it still might be useful, because each frame we used here is only 20 ms. There must be many "music-only frame" in a "music-with-singing segment." Therefore, high error rate is unavoidable, unless we are very careful when we are labeling (it's too hard to label every 20 ms ).
To solve the high-error-rate problem, we could take the majority of the classification outcome within a certain period as our final result.
張貼留言