[Discussion] reproduce evaluation table

I can't reproduce SDR value with model trained with only Musdb data
I trained model with commend lines below

spleeter train --verbose -p configs/musdb_config.json -d
spleeter evaluate -p configs/musdb_config.json -o musdb_trained_spleeter_results --mus_dir -m

and I got the result like below
and this page(https://github.com/deezer/spleeter/wiki/Separation-Performances) suggests that it should be larger then I got.
Did I miss something?

+) this is my training result
INFO:tensorflow:Saving dict for global step 200000: absolute_difference = 1.523042, bass_spectrogram = 0.19801405, drums_spectrogram = 0.35646605, global_step = 200000, loss = 1.523042, other_spectrogram = 0.56257313, vocals_spectrogram = 0.4059888
INFO:tensorflow:Saving 'checkpoint_path' summary for global step 200000: musdb_model/model.ckpt-200000
INFO:tensorflow:Loss for final step: 1.1176548.
INFO:spleeter:Model training done

Start by reproducing the commands shown with configs/musdb_config.json and the referenced MUSDB path, then compare the evaluation output with the Separation Performances wiki page. Review the attached evaluation result and the training log ending at global step 200000; done means identifying why the reproduced SDR value differs or documenting the missing training or evaluation condition.

Written by the indexing model from the issue text.