Classification et estimation de densité de microstructures triplement périodiques avec des réseaux de neurones à convolution 3D
The github repository for the article "Classification et estimation de densité de microstructures triplement périodiques avec des réseaux de neurones à convolution 3D, M. Guevara Garban, Y. Chemisky, E. Prulière, M. Clément sciencesconf.org:csma2024:501154" presented at CSMA 2024 (16ème Colloque National en Calcul de Structures).
Link to article (in French) here
Is strongly recommended to install this project using a new conda environnement
conda create -n cnn_3d python=3.11
conda activate cnn_3d
After cloning the repository, you can install the project and necessary dependencies using:
pip install .
To generate the learning database you shoud go to scripts folder and execute
python3 generate_dataset --mesh_resolution 100 --n_samples_per_class 1000 --dataset_folder dataset_mesh/
To train the CNN3D (NVIDIA GPU is required), in the same folder named scripts
execute
python3 train.py --epochs 50 --voxel_resolution 80 --dataset_folder dataset_mesh/
Results, learning curves and confusion matrix will be stored in a folder called
train_cnn, this results can be visualized using tensorboard.
tensorboard --logdir=train_cnn