About pretrained models

[sunnnnnnnny]

Hi,Thank you so much for this work.The pre-trained model of drum beat prediction has achieved good results in real samples,however i have some questions about Which data sets are these pre-trained models based on.
Hope to get your help.

[superbock]

Not sure which models you are referring to — especially since we do not have a dedicated "drum beat" prediction method. Usually the docstrings or the help of the scripts contain references to the papers, which in turn state which data it was trained with.

[sunnnnnnnny]

It is a drum transcript model pre-training model, and I did not find the paper and training data set corresponding to this pre-training model. Can you provide some help? Thank you very much.The pre-training model path is path/to/madmom-0.16.dev0\madmom\models\drums\2018\drums_crnn1_O8_S[0-2]}.pkl

[superbock]

Oh, I see. I guess packaging them was a mistake. Unfortunately, the work done here was never finalised and merged. It was done to include a method developed by a former colleague, @richard-vogl. Not sure about which paper this code actually refers to, should be one of these: https://scholar.google.com/citations?user=bxRXNs0AAAAJ

[sunnnnnnnny]

Ok.I see it,thank you for your reply.

[richard-vogl]

documentation for the drum transcription models can be found here:
https://archives.ismir.net/ismir2017/paper/000123.pdf

They are actually an adapter version for mirex, details can be found here:
http://www.music-ir.org/mirex/abstracts/2017/RV1.pdf

An old non-official madmom package with runable drum transcription code using the mirex models can be found here:
http://www.ifs.tuwien.ac.at/~vogl/models/mirex-17.zip