This is the official repository of the Dilated Regions Network (DR-Net). For technical details, please refer to:

Weakly-supervised Point Cloud Semantic Segmentation Based on Dilated Region

(1) Setup

This code has been tested with Python 3.5, Tensorflow 1.11, CUDA 9.0 and cuDNN 7.4.1 on Ubuntu 16.04/Ubuntu 18.04.

• Clone the repository

git clone --depth=1 https://github.com/LujZhang/DR-Net && cd DR-Net

• Setup python environment

conda create -n drnet python=3.5
source activate drnet
pip install -r helper_requirements.txt
sh compile_op.sh

(2) Training (S3DIS as example)

First, follow the RandLA-Net instruction to prepare the dataset.

• Start training with weakly supervised setting:

python main_S3DIS.py --mode train --gpu 0 --labeled_point 0.1%

• Evaluation:

python main_S3DIS.py --mode test --gpu 0 --labeled_point 0.1%

Acknowledge

• Our code refers to RandLA-Net and SQN, and thank a lot to these contribution.