This is a mini script to sample ModelNet40 or ShapeNetCore55v2 meshes into 3D point clouds, which are further consumed by PointNet or other set--based classification architectures. For now only random sampling is implemented.
To convert ShapeNetCore55V2 meshes into 3D point clouds with 1024 points:
python3 shapenet_obj_2_hdf5.py -i //media/daniel/HDD/ShapeNetCore_v2_orig -o /media/daniel/HDD/ShapeNetCore_v2 -s 1024
To convert ModelNet40 meshes into 3D point clouds with 1024 points:
python3 modelnet_off_2_hdf5.py -i //media/daniel/HDD/ShapeNetCore_v2_orig -o /media/daniel/HDD/ShapeNetCore_v2 -s 1024
The code is released under MIT License (see LICENSE file for details).