The official implementation of GRanD-Net
Kiran Akadas, Shankar Gangisetty
This code has been tested with Python 3.5, Tensorflow 1.11, CUDA 9.0 and cuDNN 7.4.1 on Ubuntu 16.04
Our Tensorflow code is based on RandLA-Net. For copyright license, please check code base licenses.
If you find GRandNet useful, please consider citing:
@InProceedings{Akadas_2020_ACCV,
author = {Akadas, Kiran and Gangisetty, Shankar},
title = {3D Semantic Segmentation for Large-Scale Scene Understanding},
booktitle = {Proceedings of the Asian Conference on Computer Vision (ACCV)},
month = {November},
year = {2020}
}
@article{KU202013,
title = "SHREC 2020: 3D point cloud semantic segmentation for street scenes",
journal = "Computers & Graphics",
volume = "93",
pages = "13 - 24",
year = "2020",
issn = "0097-8493",
doi = "https://doi.org/10.1016/j.cag.2020.09.006",
url = "http://www.sciencedirect.com/science/article/pii/S0097849320301400",
author = "Tao Ku and Remco C. Veltkamp and Bas Boom and David Duque-Arias and Santiago Velasco-Forero and Jean-Emmanuel Deschaud and Francois Goulette and Beatriz Marcotegui and Sebastián Ortega and Agustín Trujillo and José Pablo Suárez and José Miguel Santana and Cristian Ramírez and Kiran Akadas and Shankar Gangisetty",
keywords = "SHREC 2020, 3D point cloud, Semantic segmentation, Benchmark"
}
1)Setup python environment:
conda create -n grandnet python=3.5
source activate grandnet
pip install -r helper_requirements.txt
sh compile_op.sh
2)Place all the files to be tested in the "data_test" folder
3)Prepare dataset
cd utils/
python data_prepare.py //(Change path strings if required)
cd ..
Note: If the file contains numbers that are too large. Use the Pre_normalize.py file to normalize the numbers before running the above commands.
- Create "original_data" folder in the "data_test" folder and move all test files into it:
cd data_test/
mkdir original_data/
mv (all test_files) original_data
-
Now you have 3 folders in the "data_test" folder - original_data,original_ply,input_0.06/
-
The pre-trained model is stored in the "results" folder
-
To get the predictions for test files: (if required change the path strings in main_Semantic3D_test2.py)
python main_test.py --mode test --gpu 0 --model_path ./results/
-
The results/predictions are now stored in "test" folder.
-
To calculate mIoU :
Open calculate_miou.py and change the (Log__) in the res_path string to point to Log directory in test folder which has predictions in test folder Run
python calculate_miou.py
- Results are displayed and stored in log_MIOU.txt
Quantitative results on SHREC 2020 dataset:
Qualitative results on SHREC 2020:
Quantitative results of different approaches on S3DIS dataset:
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent