Mdx Net Save

KUIELAB-MDX-Net got the 2nd place on the Leaderboard A and the 3rd place on the Leaderboard B in the MDX-Challenge ISMIR 2021

Project README

KUIELab-MDX-Net

0. Environment

  • Ubuntu 20.04
  • at least four cuda-able GPUs (each >= 2080ti)
  • 1.5 TB disk storage for data augmentation
  • wandb for logging

Also, you must create .env file by copying .env.sample to set environmental variables.

wandb_api_key=[Your Key] # "xxxxxxxxxxxxxxxxxxxxxxxx"
data_dir=[Your Path] # "/home/ielab/repos/musdbHQ"
  • about wandb_api_key
    • we currently only support wandb for logging.
    • for wandb_api_key, visit wandb, go to setting, and then copy your api key
  • about data_dir
    • the absolute path where datasets are stored

1. Installation

conda env create -f conda_env_gpu.yaml -n mdx-net
conda activate mdx-net
pip install -r requirements.txt
sudo apt-get install soundstretch

2. Training & Submission

3. Leaderboard A vs Leaderboard B

ACKNOWLEDGEMENT

Open Source Agenda is not affiliated with "Mdx Net" Project. README Source: kuielab/mdx-net

Open Source Agenda Badge

Open Source Agenda Rating