This is official repository of FullFormer: Generating Shapes Inside Shapes paper. This work is accepted to DAGM/GCPR2023.
The project requires a Linux system that is equipped with Cuda 10.
All subsequent commands assume that you have cloned the repository in your terminal and navigated to its location.
A file named "env.yml" contains all necessary python dependencies.
To conveniently install them automatically with anaconda you can use:
conda env create -f env.yml
conda activate VQDIG
To replicate our experiments, please download the corresponding raw ShapeNet data For FullCars dataset mentioned in paper: Full Cars
For processing raw data for our model
python preprocess.py
To split the random train/validation/test split of data
python dataprocessing/create_split.py
To train autoencoder of our model
python train.py
To generate reconstruction results
python generate.py
To train the transformer to generate latent codes, which are learned during reconstruction
python training_transformer.py
To generate generation results
python latent generation.py
For questions and comments please contact Tejaswini Medi via mail