The H5 files consists of building element point clouds in 10 categories. The "data" MxNx3 matrix represents M objects, each with N points, each with x,y,z coordinates. The "label" M - length array represents an integer class label from 0 to 9.
bimnet10_train.h5:
<HDF5 dataset "data": shape (153, 2048, 3), type "<f4">
<HDF5 dataset "label": shape (153,), type "|u1">
bimnet10_test.h5
<HDF5 dataset "data": shape (47, 2048, 3), type "<f4">
<HDF5 dataset "label": shape (47,), type "|u1">
bimnet10_test_deform.h5
<HDF5 dataset "data": shape (564, 2048, 3), type "<f4">
<HDF5 dataset "label": shape (564,), type "|u1">
| Object | Count |
|---|---|
| balcony | 15 |
| beam | 15 |
| column | 30 |
| door | 31 |
| fence | 37 |
| floor | 11 |
| roof | 21 |
| stairs | 15 |
| wall | 24 |
| window | 17 |
| total | 216 |
pcl_viewer door.pcd
python deform_pcd.py door.pcd out.pcd --rotate; pcl_viewer out.pcd
python deform_pcd.py door.pcd out.pcd --noise; pcl_viewer out.pcd
python deform_pcd.py door.pcd out.pcd --bend; pcl_viewer out.pcd
python deform_pcd.py door.pcd out.pcd --truncate; pcl_viewer out.pcd
python deform_pcd.py door.pcd out.pcd --rotate --noise --bend --truncate; pcl_viewer out.pcd
Convert CAD model (PLY file) to point cloud (PCD file):
python mesh2pcd.py door.ply door.pcd
Convert point clouds from H5 file to folder of PCD files:
python h52pcd.py bimnet10_train.h5 pcd/
Convert folder of point clouds in PCD format to H5 file:
python pcd2h5.py pcd/ bimnet10_train.h5
Visualize point cloud data (requires Point Cloud Library)
pcl_viewer pcd/0-cloud.pcd
Train / Test Network:
python bimnet.py
@inproceedings{ChenCho2018,
author = "Jingdao Chen and Yong K Cho and Jun Ueda",
booktitle = {Proceedings of the 2018 IEEE Conference on Robotics and Automation (ICRA)},
title = "Sampled-Point Network for Classification of Deformed Building Element Point Clouds",
year = {2018}
}