results when replicate the code following Onsets and Frames: Dual-Objective Piano Transcription

[Dream-High]

Hello, @jongwook, Thanks for your opening codes.
Recently, I use your code in order to get the performance in paper "Onsets and Frames: Dual-Objective Piano Transcription". I use the MAPS to train the model and the batch_size=4, iteration= 358000. When evaluating, I get the performance as following.

Some metrics appear to be quite low, especially the frame metrics which are 82.2/70.4/75.5 whereas the "Onsets and Frames: Dual-Objective Piano Transcription" paper reports 88.53/70.89/78.3

Do you know the reasons about that?
Thanks a lot

[xk-wang]

the magenta team trained the onsets and frames model using maestro dataset instead of maps dataset.

[Dream-High]

@xk-wang There are three papers using the onsets and frames model, while in the "Onsets and Frames: Dual-Objective Piano Transcription", the model is trained on MAPS datasets

[xk-wang]

@Dream-High I also used this code, it is not completely the same with the original onsets and frames model. You should use the original Tensorflow version code and convert it to PyTorch yourself. I think this code just implements the main idea, but some details are missing comprared with the Tensorflow version.