TriVol: Point Cloud Rendering via Triple Volumes

This is the source code of "TriVol: Point Cloud Rendering via Triple Volumes" (CVPR-2023).

Requirements (Tested on 4 * RTX3090)

• Linux
• Python == 3.8
• Pytorch == 1.13.0
• CUDA == 11.7

Installation

Install from environment.yml

You can directly install the requirements through:

$ conda env create -f environment.yml

Or install packages seperately

• Create Environment

  $ conda create --name trivol python=3.8
  $ conda activate trivol

• Pytorch (Please first check your cuda version)

  $ conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia

• NerfAcc

  $ pip install git+https://github.com/KAIR-BAIR/nerfacc.git

• torch_scatter

  pip install torch-scatter -f https://data.pyg.org/whl/torch-1.13.0+cu117.html

• Other python packages: open3d, opencv-python, etc.

Training

Training on the ShapeNet Car dataset

• Dataset: Download our processed ShapeNet Car from HuggingFace.

• Run: we recommand you run the code in multi-gpu mode.

  $ conda activate trivol
  $ python train.py \
          --scene_dir /path/to/shapenet \
          --dataset shapenet \
          --val_mode val \
          --max_epochs 200 \
          --lr 0.0005 \
          --batch_size 1 \
          --ngpus 4 \
          --feat_dim 16 \
          --patch_size 64 \
          --exp_name shapenet_car

• example results

  

Training on the ScanNet dataset

• Dataset: Download and extract data from original ScanNet-V2 preprocess.

  Dataset structure:

  ── scannet
    └── scene0000_00
        ├── pose
              └──1.txt
         ├── intrinsic
              └──*.txt
        ├── color
               └──1.jpg
        └── scene0000_00_vh_clean_2.ply
        └── images.txt
    └── scene0000_01
  

• Run: we recommand you run the code in multi-gpu mode.

  $ conda activate trivol
  $ python train.py \
          --scene_dir /path/to/scannet \
          --dataset scannet \
          --val_mode val \
          --max_epochs 500 \
          --lr 0.0005 \
          --batch_size 1 \
          --ngpus 4 \
          --feat_dim 16 \
          --patch_size 32 \
          --exp_name scannet

• example results

  

Contact

Tao Hu - [email protected]

Citation

@InProceedings{Hu_2023_CVPR,
    author    = {Hu, Tao and Xu, Xiaogang and Chu, Ruihang and Jia, Jiaya},
    title     = {TriVol: Point Cloud Rendering via Triple Volumes},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {20732-20741}
}