3D Point Cloud Data Augmentation via Scene Representation Network

Report | Slides

Teammate

Pei-Tse Chiang, Meng-Hsun Tsai, Zhi-Yi Chin, Chieh-Ming Jiang

Introduction

We design a 3D point cloud augmentation based on a novel view synthesis method, SRN. The 3D point cloud is a set of data points in 3D space, consisting of coordinates in world space and RGB color information. However, the precision is limited if there are insufficient points to describe the objects. Moreover, the amount of point cloud training samples is also limited, leading to the poor diversity of data. This motivates us to design a data augmentation method for the point cloud. As our base model, given extrinsic and intrinsic matrix, the SRN model can estimate the world coordinate and get the corresponding depth and RGB color information, which can easily be transferred to point cloud form. After the SRN model is trained, we can render new images of unseen views by changing the extrinsic matrix and consequently realize the point cloud augmentation. To verify whether the augmented point cloud is good enough to describe the 3D objects, we use PointNet as our evaluation model to evaluate the quality of the augmentation.

Getting the code

You can download a copy of all the files in this repository by cloning this repository:

git clone https://github.com/joycenerd/image-super-resolution.git

Requirements

You need to have Anaconda or Miniconda already installed in your environment. To install requirements:

conda env create -f environment.yml

Data

Download raw data

• ModelNet
• ShapeNet

Data conversion

• convert .off to .obj:

  cd model-converter-python
  python convert.py \
  --data-root <data_dir> \
  --output-root <save_dir>
  

• Render 50 views (rgb, poses)

  Please refer to here for detail.

• convert .obj to .pcd

  Please refer to obj2pcd for detail

• convert .off to .ply

  cd shapenet_renderer
  python off2ply.py --data-root <data_dir> --output-root <save_dir>
  

• convert .ply to .pcd You can use open3d api for conversion

  import open3d as o3d
  pcd = o3d.io.read_point_cloud("XXX.ply")
  o3d.io.write_point_cloud("XXX.ply", pcd)
  

Train SRN

1. generate training and testing path

   cd scene-representation-networks
   python choose_train_test_data.py --ratio 0.7 --class_name airplane
   

2. Training

   cd scene-representation-networks
   python train.py --config_filepath train_configs/modelnet_all.yml --data_root train_data_path.txt --logging_root <save_dir> --ratio 0.7 --gpu 0 --train_class airplane
   

Test SRN

1. Testing

   cd scene-representation-networks
   python test.py --config_filepath test_configs/modelnet_all.yml --data_root test_data_path.txt --logging_root <save_dir> --checkpoint <ckpt_path> --ratio 0.7 --gpu cuda:0 --train_class airplane --class_name airplane
   

2. Create point cloud points

   cd scene-representation-networks
   python create_pcd_MN.py --ratio 0.X --iter X --class_name XXX
   

Train PointNet

cd PointNet
python train_cls.py --output_dir /eva_data_0/augment_output_single_version/PointNet/XXX/real_0.X/ --data-txt /eva_data_0/augment_output_single_version/ratio_data/XXX/0.X/train_data_path.txt --augment_data_dir /eva_data_0/augment_output_single_version/3D_points/XXX/real_0.X/ --gpu X

Test PointNet

cd PointNet
python test_cls.py --output_dir <npy_files_dir> --gpu 0

Results

GitHub Acknowledge

We thank the authors of these repositories:

• vsitzmann/scene-representation-networks
• yanx27/Pointnet_Pointnet2_pytorch
• tforgione/model-converter-python
• jiangwei221/obj2pcd

Citation

If you find our work useful in your project, please cite:

@misc{
    title = {3D Point Cloud Data Augmentation via Scene Representation Network},
    author = {Pei-Tse Chiang, Meng-Hsun Tsai, Zhi-Yi Chin, Chieh-Ming Jiang},
    url = {https://github.com/joycenerd/3D_Augmentation},
    year = {2022}
}

Contributing

If you'd like to contribute, or have any suggestions, you can contact us at [email protected] or open an issue on this GitHub repository.

All contributions welcome! All content in this repository is licensed under the MIT license.