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

see README_SUBMISSION.md

ACKNOWLEDGEMENT

• This repository is based on Lightning-Hydra Template
• Also, facebook/demucs