Project Website: https://hyperplane-lab.github.io/Generative-3D-Part-Assembly/ Base code is at https://github.com/hyperplane-lab/Generative-3D-Part-Assembly
1.) Put their code into /src/{their repo}
Delete their .git file and check it into our repo
2.) Get their pretrained model / dataset
Dataset is http://download.cs.stanford.edu/orion/genpartass/prepare_data.zip
Pretrained model is http://download.cs.stanford.edu/orion/genpartass/checkpoints.zip put the zip in the exps file not in the prep_data area
when unzipped it should be at /exps/prepare_data and then a bunch of npy files (download using firefox if using linux for some reason chromium download is corrupted)
run conda env create -f environment.yaml . activate PartAssembly cd exps/utils/cd python setup.py build
if it breaks you may need to run python setup.py install make sure cudatoolkit is installed and the PATH is set properly
Might install some other things like scikit and scikitlearn as well as ipdb in the venv
their github says dynamic_graph_learning that's a typo
to train cd exps/Our_Method-dynamic_graph_learning/scripts/ ./train_dynamic.sh
to test cd exps/Our_Method-dynamic_graph_learning/scripts/ ./test_dynamic.sh
replace Our_Method-dynamic_graph_learning with the folder so Baseline-LSTM etc.
3.) Run their experiments (B-LSTM, B-Global, and their real one)
B-Complement would be cool, but might be hard to do
4.) Put any custom code in /src/ outside of their folder so we can separate their code from ours.