Reproduce the paper on MIR_ST500 or MDB-stem-synth

[hdnhan]

Hi, I'm new here and I wanted to reproduce the result from the paper. Currently I can reproduce the result on MIR-1K, the default config matches the meta data of MIR-1K. However, I have troubles with the other two datasets:

1. Now I want to try MDB-stem-synth. As far as I know, the hop size of the annotation is ~2.8ms, but your config is 20ms. What did you do to get a frame of 20ms for pitch note for this dataset?
2. The same question for MIR_ST500 which is a real dataset, not synthetic one.

Thanks for the contribution.

[Joanna1212]

same question here