Symbolic-Music-Genre-Transfer-with-CycleGAN-for-pytorch

7.30 Fixed bug ,added to_binary function

12.25 Polish

If you find any bugs, please contact me!!!

Introduction

This repository is based on the paper Symbolic Music Genre Transfer with CycleGAN and the repository CycleGAN-Music-Style-Transfer-Refactorization

Installation

• Clone this repo:

  git clone https://github.com/Asthestarsfalll/Symbolic-Music-Genre-Transfer-with-CycleGAN-for-pytorch.git 
  cd Symbolic-Music-Genre-Transfer-with-CycleGAN-for-pytorch

• Install dependencies:

  pip install -r requirements.txt

Usage

Prepare the dataset

Download the dataset which can be use directly

Put them into the folder traindata, make sure they're like this:

-- Symbolic-Music-Genre-Transfer-with-CycleGAN-for-pytorch

	-- traindata

		|- JCP_mixed

		|- CP_P

		|- CP_C

		|- JP_P

		|- JP_J

		|- JC_C

		|- JC_J

For those who want to use their own datasets, or want to try all the preprocessing steps, please take a look at Testfile.py and convert_clean.py files which are in this repo. There are 6 steps in the Testfile.py which are remarked clearly. You can just comment/uncomment and run these code blocks in order. The second step is to run convert_clean.py. Please make sure all the directory paths have correct settings though it might take some time! It's very important! In addition, please ensure there are the same number of phrases (always downsampling) for each genre in a genre pair for the sake of avoiding imbalance. E.g., for classic vs. jazz, you might get 1000 phrases for classic and 500 phrases for jazz, then you should downsample the phrases for classic to 500.

Train

use sh train.sh to simply train the model.

or

python train.py --help
python train.py --epoch 5 --batch-size 2 --model-name CP --data-mode full

I found that the model in the later stages of training produced almost the same test results.I think the Generator learned the discriminant mode of Discriminator.So I suggest that try to increase the weight of cycle loss and mixed loss( which makes the Discriminator learn more discriminant mode ).

You will get the .pth model file in the folder saved_models/model_name

Test

python test.py --data-dir path/to/data --model-dir path/to/model --batch-size 2 --model-name CP --test-mode A2B

The model have the mode A2B and B2A for testing, which are supposed to Indicate the first music genre to second music genre.

You will get the test results in folder test.