Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds

Bo Yang, Jianan Wang, Ronald Clark, Qingyong Hu, Sen Wang, Andrew Markham, Niki Trigoni. arXiv:1906.01140, 2019.

(1) Setup

ubuntu 16.04 + cuda 8.0

python 2.7 or 3.6

tensorflow 1.2 or 1.4

scipy 1.3

h5py 2.9

open3d-python 0.3.0

Compile tf_ops

(1) To find tensorflow include path and library paths:

import tensorflow as tf
print(tf.sysconfig.get_include())
print(tf.sysconfig.get_lib())

(2) To change the path in all the complie files, e.g. tf_ops/sampling/tf_sampling_compile.sh, and then compile:

cd tf_ops/sampling
chmod +x tf_sampling_compile.sh
./tf_sampling_compile.sh

(2) Data

S3DIS: https://drive.google.com/open?id=1hOsoOqOWKSZIgAZLu2JmOb_U8zdR04v0

百度盘: https://pan.baidu.com/s/1ww_Fs2D9h7_bA2HfNIa2ig 密码:qpt7

Acknowledgement: we use the same data released by JSIS3D.

(3) Train/test

python main_train.py

python main_eval.py

(4) Quantitative Results on ScanNet

(5) Qualitative Results on ScanNet

More results of ScanNet validation split are available at: More ScanNet Results

To visualize: python helper_data_scannet.py

(6) Qualitative Results on S3DIS

(7) Training Curves on S3DIS

(8) Video Demo (Youtube)

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{yang2019learning,
  title={Learning object bounding boxes for 3d instance segmentation on point clouds},
  author={Yang, Bo and Wang, Jianan and Clark, Ronald and Hu, Qingyong and Wang, Sen and Markham, Andrew and Trigoni, Niki},
  booktitle={Advances in Neural Information Processing Systems},
  pages={6737--6746},
  year={2019}
}

Related Repos

1. RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds
2. SoTA-Point-Cloud: Deep Learning for 3D Point Clouds: A Survey
3. SensatUrban: Learning Semantics from Urban-Scale Photogrammetric Point Clouds
4. SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration