# RandLA-Net-pytorch

**Repository Path**: pisceszove/RandLA-Net-pytorch

## Basic Information

- **Project Name**: RandLA-Net-pytorch
- **Description**: 点云处理网络2020-cvpr oral
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 1
- **Created**: 2021-05-06
- **Last Updated**: 2021-05-06

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds

This repository contains a PyTorch implementation of [RandLA-Net](http://arxiv.org/abs/1911.11236).

## Preparation

1. Clone this repository

   ```sh
   git clone https://github.com/aRI0U/RandLA-Net-pytorch.git
   ```

2. Install all Python dependencies

  ```sh
    cd RandLA-Net-pytorch
    pip install -r requirements
  ```

***Common issue***: *the setup file from `torch-points-kernels` package needs PyTorch to be previously installed. You may thus need to install PyTorch first and then torch-points-kernels.*

4. Download a dataset and prepare it. We conducted experiments with [Semantic3D](http://www.semantic3d.net/) and [S3DIS](http://buildingparser.stanford.edu/dataset.html).

  To setup Semantic3D:

   ```sh
   cd RandLA-Net-pytorch/utils
   ./download_semantic3d.sh
   python3 prepare_semantic3d.py # Very slow operation
   ```

   To setup SDIS, register and then download the `zip` archive containing the files [here](http://buildingparser.stanford.edu/dataset.html#Download). We used the archive which contains only the 3D point clouds with ground truth annotations.

   Assuming that the archive is located in folder `RandLA-Net-pytorch/datasets`, then run:

   ```sh
   cd RandLA-Net-pytorch/utils
   python3 prepare_s3dis.py
   ```

5. Finally, in order to subsample the point clouds using a grid subsampling, run:
  ```sh
  cd RandLA-Net-pytorch/utils/cpp_wrappers
  ./compile_wrappers.sh   # you might need to chmod +x before
  cd ..
  python3 subsample_data.py
  ```


## Usage

- Train a model

  ```sh
  python3 train.py
  ```

  A lot of options can be configured through command-line arguments. Type `python3 train.py --help` for more details.

- Evaluate a model

  ```sh
  python3 test.py
  ```

### Visualization

One can visualize the evolution of the loss with Tensorboard.

On a separate terminal, launch:

  ```sh
  tensorboard --logdir runs
  ```

## Citation

This work implements the work presented in [RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds](http://arxiv.org/abs/1911.11236).

The original implementation (in TensorFlow 1) can be found [here](https://github.com/QingyongHu/RandLA-Net).

To cite the original paper:
  ```
  @article{RandLA-Net,
    arxivId = {1911.11236},
    author = {Hu, Qingyong and Yang, Bo and Xie, Linhai and Rosa, Stefano and Guo, Yulan and Wang, Zhihua and Trigoni, Niki and Markham, Andrew},
    eprint = {1911.11236},
    title = {{RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds}},
    url = {http://arxiv.org/abs/1911.11236},
    year = {2019}
  }
  ```
<!--

## TODOs

make data.py support different input dims

optimization:
- limit memory usage
- see whether cross entropy with dtype uint8 possible
- make num_workers work -->

## Warning

*This repository is still on update, and the segmentation results we reach with our implementation are for now not as good as the ones obtained by the original paper.*