This project aims to reconstruct surfaces from point clouds, generate a mesh, and visualize it using 3D viewers such as MeshLab.

First, create an Anaconda environment called conv_onet using the provided environment.yaml file.

conda env create -f environment.yaml
conda activate conv_onet

Next, compile the extension modules required for the project.

python setup.py build_ext --inplace

1. Place your raw data in the data/raw folder.
2. Place your .npz data into the data folder, such as data/daisue/raw folder. The .npz file should contain a points variable storing each point in an (x, y, z) format.

To process your raw data by sampling and creating normal vectors, run the following bash script:

bash scripts/dataset_daisue/run_build_dataset.sh

After running the script, a new .npz file with the normals vector calculated will be generated. You can check the code in scripts/dataset_daisue/build_dataset.py for more details.

To reconstruct the surface using a pretrained model trained on Matterport3D, run the following command:

python generate.py configs/pointcloud_crop/daisue.yaml

After generating the mesh, you can view it using 3D viewers such as MeshLab.

This project is modified from Convolutional Occupancy Networks.

This project is licensed under the terms of the MIT license.