KOREGRAPH

An IA for music driven choregraphy generation.

---

Technos : Python, Poetry, Tensorflow, Keras, Google Cloud

Author : Anatole-DC

---

Summary

• Summary
• Requirements
  • Git
  • Python 3.8.10
  • Poetry
  • VSCode (Optional)
• Install the project
  • Clone the repository
  • Install the dependencies
• Get a prediction
• Scripts
• Ideas
• License

Requirements

Git

git --version
# git version 2.34.1

Python 3.8.10

python --version
# Python 3.10.12

Poetry

poetry run --version
# Poetry (version 1.6.1)

If you are using pyenv

Activate koregraph

pyenv local koregraph

Install poetry

pip intall poetry

VSCode (Optional)

This template is configured to work with the VSCode editor, but it does not required it to be used.

Install the project

Clone the repository

git clone https://github.com/Anatole-DC/koregraph
cd koregraph

Install the dependencies

poetry shell
poetry install

In dev mode :

poetry install --with dev,viewer

Get a prediction

Generate features and output pkl files

poetry run generate

Train a model

poetry run train -m model_kelly

Get a video

poetry run predict -m model_kelly -a mBR0

Scripts

Export presentation (github pages)

poetry run jupyter nbconvert --no-input frontend/views/presentation.ipynb  --to slides --stdout > documentation/pages/index.html

Train a model

poetry run train --help

# Pass the -d option to only take a random sample of the dataset
poetry run train -m modelkelly -d 0.5

# Train on the cloud (ensure the environment variable are set)
poetry run train -m modelkelly --with-cloud -d

Predict a choregraphy from an audio

poetry run predict -a mBR1 -m model -i 01

Build viewer video from choregraphy

poetry run viewer -c data/keypoints2d/gBR_sBM_cAll_d04_mBR0_ch01.pkl  # Path to your video

Create 5 sec chunks for a choreography

poetry run chunk -c data/keypoints2d/gBR_sBM_cAll_d04_mBR0_ch02.pkl -s 5

If the music already exists, but you want to split it again for some reason, add the following parameter --reload-music

Plot the loss

poetry run evaluate -m model_name

If history is a buckup file, add -b

Ideas

Pitch en ligne

Transformer mercredi deuxième semaine

License

Shield:

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.