VAE implicit parameterized 3d structure dataset represented by signed distance field(SDF-VAE)

prepared the data

Each cell in the dataset is saved as a .npy named according to its index(e.g.cell_index.npy). These files are located at:

./src/data/npy_file

Initial Setup

1. Ensure the configuration settings are properly set in pre_dataset_config.json
2. Use dataset.py to read all SDF files and prepare the training and test datasets.

Train the model

1. Ensure the configuration settings of the trainig model config_SDF_VAE.json
2. The structure of the Convolutional Neural Network (CNN) depends on the resolution of the SDFs. By default, the resolution is set to $50^3$, If you change the resolution, you must also adjust the parameters of the last layers to guarantee the same output resolution.
3. hight recomend GPU mahinve to run the trainig code main_train_sdf_vae.py, after training, reuse the model by reuse_model.py

Display the reconstruct surface by decoder

View STL model