Vis2Mesh

This is the offical repository of the paper:

Vis2Mesh: Efficient Mesh Reconstruction from Unstructured Point Clouds of Large Scenes with Learned Virtual View Visibility

ICCV | Arxiv | Presentation

@InProceedings{Song_2021_ICCV,
    author    = {Song, Shuang and Cui, Zhaopeng and Qin, Rongjun},
    title     = {Vis2Mesh: Efficient Mesh Reconstruction From Unstructured Point Clouds of Large Scenes With Learned Virtual View Visibility},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {6514-6524}
}

Updates

2021/9/6: Intialize all in one project. Only this version only supports inferencing with our pre-trained weights. We will release Dockerfile to relief deploy efforts.

TODO

Ground truth generation and network training.
Evaluation scripts

Build With Docker (Recommended)

Install nvidia-docker2

# Add the package repositories
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker

Build docker image

docker build . -t vis2mesh

Build on Ubuntu

Please create a conda environment with pytorch and check out our setup script:

./setup_tools.sh

Usage

Get pretrained weights and examples

pip install gdown
./checkpoints/get_pretrained.sh
./example/get_example.sh

Run example

The main command for surface reconstruction, the result will be copied as $(CLOUDFILE)_vis2mesh.ply.

python inference.py example/example1.ply --cam cam0

We suggested to use docker, either in interactive mode or single shot mode.

xhost +
name=vis2mesh
# Run in interactive mode
docker run -it \
--mount type=bind,source="$PWD/checkpoints",target=/workspace/checkpoints \
--mount type=bind,source="$PWD/example",target=/workspace/example \
--privileged \
--net=host \
-e NVIDIA_DRIVER_CAPABILITIES=all \
-e DISPLAY=unix$DISPLAY \
-v $XAUTH:/root/.Xauthority \
-v /tmp/.X11-unix:/tmp/.X11-unix:rw \
--device=/dev/dri \
--gpus all $name

cd /workspace
python inference.py example/example1.ply --cam cam0

# Run with single shot call
docker run \
--mount type=bind,source="$PWD/checkpoints",target=/workspace/checkpoints \
--mount type=bind,source="$PWD/example",target=/workspace/example \
--privileged \
--net=host \
-e NVIDIA_DRIVER_CAPABILITIES=all \
-e DISPLAY=unix$DISPLAY \
-v $XAUTH:/root/.Xauthority \
-v /tmp/.X11-unix:/tmp/.X11-unix:rw \
--device=/dev/dri \
--gpus all $name \
/workspace/inference.py example/example1.ply --cam cam0

Run with Customize Views

python inference.py example/example1.ply Run the command without --cam flag, you can add virtual views interactively with the following GUI. Your views will be recorded in example/example1.ply_WORK/cam*.json.

Main View

Navigate in 3D viewer and click key [Space] to record current view. Click key [Q] to close the window and continue meshing process.

Record Virtual Views

GDAOSU / vis2mesh

readme