Which dataset is used to train the model?

[mtalhabalci]

Hello, thanks this repository

I am willing to train a model with a different music notation (Turkish music notation) rather than european. There are some microtonic symbols in my case.

I want to make some small proof of concept before going further. I am observing the reporitory and did not manage to see a dataset.

And the question is which dataset is used to train this model?

[BreezeWhite]

The dataset used is stated under README -> Technical Details -> Model Training.

[mtalhabalci]

Hi again,

For u_net and seg_net models, I am fine but here's a question:

To identify invidual symbol types on the predictions, SVM models are used. The data used to train SVM models are extracted from DeepScores-extended.

In readme you mention about the dataset used for training SVM models. Can we reach those datasets?

Or can you ellaborate on how did you extract the data from the ds2 dataset?

Asking because I need to train some models to get to know the new symbols and I do not want to change the structure you have provided.

Thanks indeed.

[BreezeWhite]

You can find the data collection function in classifier.py:L33-L68. After the collection, manually filter out samples that are overly transformed that hardly maintain the original shape. Finally use these samples to train the SVM models using the train_* functions in the same file.