Skip to content

hojoonson/3d_retrieval_with_pointnet

Repository files navigation

3d_retrieval_with_pointnet

References

PointNet Source : https://github.com/charlesq34/pointnet

ShapeNet Data : https://shapenet.org/

Citation

Paper : https://link.springer.com/article/10.1007/s12206-021-1024-z

@article{son2021three,
  title={Three-dimensional model retrieval in single category geometry using local ontology created by object part segmentation through deep neural network},
  author={Son, Hojoon and Lee, Soo-Hong},
  journal={Journal of Mechanical Science and Technology},
  volume={35},
  number={11},
  pages={5071--5079},
  year={2021},
  publisher={Springer}
}

Train with Tensorflow Docker

Tf1.15 with Python 2

nvidia-docker run -v {Your Project Path}:/projects -it tensorflow/tensorflow:1.15.0-gpu
cd part_seg
python train.py

How to use

Use Python3. You should have full ShapeNet Data.

1. extract aircrat information

python extract_aircraft.py

2. create json file

airplane_before_ontology.json is result file.

python create_aircraft_json.py

3. create ontology

python create_aircraft_ontology.py

4. Calculate similarities

Name_all.csv is manually created dbpedia mapping data.

cd retrieval
python rank_similarity.py