Hello.
I try to reproduce the paper results by using the GitHub script.
The model, train, test code is the same as the explanation in the paper.
But, The data used in this experiment was not complete data.
Without the process of making the formal data set(SHS100K, youtube, cover 80) you used, the program was not worked correctly as shown in below figure.
I'd like to show the same performance by using the data you used.
Could you share preprocessing code for data?
I know that the 'gencqt.py' code generates cqt from audio.
Also, Due to the dataset size is too big, you cannot directly offer the data.
So, please offer the code to download the 'SHS100K_url' or the 'npz' files that compressed the file.
Also, how can I get the data of 'songs80_list' and 'Mazurkas_list'?
Thank you for your appreciation.
Hello. I try to reproduce the paper results by using the GitHub script.
The model, train, test code is the same as the explanation in the paper. But, The data used in this experiment was not complete data. Without the process of making the formal data set(SHS100K, youtube, cover 80) you used, the program was not worked correctly as shown in below figure.
I'd like to show the same performance by using the data you used. Could you share preprocessing code for data?
I know that the 'gencqt.py' code generates cqt from audio. Also, Due to the dataset size is too big, you cannot directly offer the data. So, please offer the code to download the 'SHS100K_url' or the 'npz' files that compressed the file.
Also, how can I get the data of 'songs80_list' and 'Mazurkas_list'? Thank you for your appreciation.