Point Cloud Super Resolution with Adversarial Residual Graph Networks

[arXiv] [Home]

Official implementation of Point Cloud Super Resolution with Adversarial Residual Graph Networks.

@inproceedings{wu2019point,
  title     = {Point Cloud Super Resolution with Adversarial Residual Graph Networks},
  author    = {Wu, Huikai and Zhang, Junge and Huang, Kaiqi},
  booktitle = {arXiv preprint arXiv:1908.02111},
  year      = {2019}
}

Contact: Hui-Kai Wu (huikaiwu@icloud.com)

Install

The code is tested with TF1.5 (higher version should also work) and Python 3.5 on Ubuntu 16.04

1. Clone the repository:

   git clone https://github.com/wuhuikai/PointCloudSuperResolution
   cd PointCloudSuperResolution

2. Install Requirements

   pip install -r requirements.txt

3. Compile the TF operators

   cd code/tf_ops/CD && CUDA_HOME=[CUDA_HOME] bash tf_nndistance_compile.sh
   cd code/tf_ops/grouping && CUDA_HOME=[CUDA_HOME] bash tf_grouping_compile.sh
   cd code/tf_ops/sampling && CUDA_HOME=[CUDA_HOME] bash tf_sampling_compile.sh

Train & Test

1. Download the training patches in HDF5 format from GoogleDrive and put it in folder data.

2. Train [Optional]

   cd code
   python main_gan.py --phase train --dataset ../data/Patches_noHole_and_collected.h5
   python main_gan.py --phase train --dataset ../data/Patches_noHole_and_collected.h5 --gan --log_dir ../model/model_res_mesh_pool_gan_ft --batch_size 16 --model_path ../model/model_res_mesh_pool/model-80 --max_epoch 40

3. Predict

   python main_gan.py --dataset ../data/test_data/our_collected_data/input --log_dir ../model/model_res_mesh_pool_gan_ft

4. Evluation

   cd evaluation_code
   conda install cgal
   mkdir build && cd build && cmake .. && make && mv evaluation ../ && cd .. && rm -rf build
   
   python evaluation_cd.py --pre_path ../model/model_res_mesh_pool_gan_ft/result/input --gt_path ../data/test_data/our_collected_data/gt
   python evaluation.py --pre_path ../model/model_res_mesh_pool_gan_ft/result/input --gt_path ../data/test_data/our_collected_data/gt_off --save_path ../model/model_res_mesh_pool_gan_ft/result/input_nuc

Acknowledgement

The code is modified from PointNet++ and PU-Net.