This repository provides Tensorflow implementations for Rank3DGAN paper.
This code is heavily based on Multi-chart Generative Surface Modeling implementation multichart3dgans.
shape_version
: the project code for watertight meshes (human and bird shapes) that requires triplet of landmarks
face_version
: the project code for face meshes that requires quadriplet of landmarks
For all the details of the data prepration, please check the paper Appendix.
For the data pre/post-processing, please check matlab folder.
- Python 3.5
- TensorFlow 1.6
bunch
andtqdm
packages
Check matlab/createDataset.m
for the details. To transform the charts to tfrecords format use:
- For the case of single meshes dataset:
python3 convert_to_tfrecords.py --database_signature=<database_signature>
- For the case of pairwise comparison meshes dataset:
python3 convert_to_tfrecords.py --database_signature=<database_signature> -p
python3 gan_main.py -c=configs/<my_config>.json
python3 evaluate_main.py -c=configs/config.json
After getting the generated charts. Use matlab/inspectGeneratedData.m
to obtain the resulting meshes.