HMG: Human Motion Generation

Generating diverse Rehabilitation human motions with a score.

Installation 👷

Create conda environment

Instructions

conda create python=3.10 --name py310
conda activate py310

Install the following packages:

pip install torchvisison = 0.17
pip install numpy
pip install imageio
pip install pandas
pip install seaborn
pip install matplotlib
pip install scikit-learn

The code was tested on Python 3.10.13 and PyTorch 2.2

How to Train the network:

The command to launch a training experiment is the folowing:

python3 train_vae.py --generative-model CVAE --dataset Kimore --output-directory results/ --runs 5 --weight-rec 0.9 --weight-kl 1e-3 --epochs 2000 --device cuda

Model

• generative-model=CVAE: select which generative model to use to generate samples.

runs

• runs = int: number of times to train the model.

Device

• Device=cuda: training with CUDA, on an automatically selected GPU (default).
• Device=mps: training with MPS, training on GPU for MacOS devices with Metal programming framework.
• Device=cpu: training on the CPU.

How to generate skeletons:

The command to launch a generation experiment is the folowing:

python3 generate_samples.py --generative-model CVAE --dataset Kimore --output-directory results/ --device cuda --class_index 0

class_index

• class_index = int: chich class you want to generate from.

Evaluation with the metrics: