Concise Plane Arrangements for Low-Poly Surface and Volume Modelling

This repository contains the official implementation of the ECCV 2024 paper "Concise Plane Arrangements for Low-Poly Surface and Volume Modelling". Given a set of input planes and corresponding inlier points the resulting plane arrangement can be used for polygon surface mesh reconstruction and convex decomposition of volumes.

📋 Features

Reading of vertex groups (.vg, .npz) as input (e.g. from here)
Fast and memory efficient concise plane arrangement construction
Storing of the arrangement as a binary space partitioning tree (BSP-tree)
Interior / exterior labelling of the arrangement cells using point normals or a reference mesh
Further simplification of the arrangement based on a careful analysis of the BSP-tree
Extraction of a concise convex decomposition (i.e. interior cells of the arrangement)
Extraction of a concise polygon surface mesh (i.e. interface polygons between interior and exterior cells).

🧱 Installation

Simply clone the repository and install in a new conda environment using pip:

git clone https://github.com/raphaelsulzer/compod.git
cd compod
sudo apt-get update && sudo apt-get install libgomp1 ffmpeg libsm6 libxext6 -y
bash install.sh    # this step may take some time

You are now ready to use COMPOD. You can test your installation with:

conda activate compod
cd example
python example.py

COMPOSE

COMPOSE is an extension for COMPOD that implements some routines for Surface Extraction in C++. Those are:

a fast interior / exterior labelling of the arrangement cells based on interior point tests using a reference mesh.
a simplification of the extracted surface mesh based on a Constrained Delaunay Triangulation of the corner vertices of each planar region.

To install COMPOSE you need to follow the steps below:

cd compose
conda install -y conda-forge::spdlog conda-forge::cgal anaconda::mpfr 
pip install .

💻 Usage

from pycompod import VertexGroup, PolyhedralComplex

model = "anchor"

file = "data/{}/convexes_refined/file.npz".format(model)
vg = VertexGroup(file,verbosity=20)
cc = PolyhedralComplex(vg,device='gpu',verbosity=20)

cc.construct_partition()
cc.add_bounding_box_planes()
cc.label_partition(mode="normals")
# ## needs compose extension
# cc.label_partition(mode="mesh",mesh_file="data/{}/surface/dense_mesh.off".format(model))

cc.simplify_partition_tree_based()
cc.save_partition("data/{}/partition/tree_simplified_partition.ply".format(model), export_boundary=True)
cc.save_partition_to_pickle("data/{}/partition".format(model))

cc.save_surface(out_file="data/{}/surface/complex_mesh.obj".format(model), triangulate=False)
## needs compose extension
cc.save_simplified_surface(out_file="data/{}/surface/polygon_mesh.obj".format(model), triangulate=False)
cc.save_simplified_surface(out_file="data/{}/surface/triangle_mesh.obj".format(model), triangulate=True)  
cc.save_wireframe(out_file="data/{}/surface/wireframe.obj".format(model))

📸 Examples

Please see the example/ folder.

📖 Citation

If you use this work please consider citing:

@misc{sulzer2024concise,
      title={Concise Plane Arrangements for Low-Poly Surface and Volume Modelling}, 
      author={Raphael Sulzer and Florent Lafarge},
      year={2024},
      eprint={2404.06154},
      archivePrefix={arXiv},
      primaryClass={cs.CG}
}