3D Mesh Decimation Library Comparison

Key Takeaways

In case you experience a lack of time to scan the entire research, we place this abstract at the very beginning, offering a swift overview. For projects that demand both speed and accuracy, MeshLib (MT and ST) consistently excels—combining quick processing times with robust mesh quality under heavy and moderate simplifications. If absolute geometric precision overrides speed concerns, Rhino 3D stands out for its higher fidelity (especially by Hausdorff and Average Absolute Distance metrics), albeit with slower execution. Meanwhile, CGAL could be a solid option for tasks prioritizing visual fidelity, despite its lower speed and marginally lower metrics. Finally, VTK and MeshLab Clustering Decimation should generally be avoided whenever high reliability and quality are crucial, as they frequently introduce structural issues and high error rates.

Introduction

The idea behind mesh simplification (also known as mesh decimation) is reducing the number of polygons in a given 3D mesh while preserving its overall shape and visual fidelity. Such a technique is understandably essential for optimizing 3D models. Namely, the goal here is to make meshes more suitable for real-world applications, which demand:

Faster rendering and economical memory usage (especially in resource-constrained environments);
Cost-efficiency which is to be secured thanks to minimized processing times and slightly reduced material consumption (say, like 3D printing);
Compatibility based on adapting complex models for utilization on a variety of platforms.

Thus, from 3D printing and architectural visualization to medical imaging and geospatial modeling, mesh simplification underpins a wide range of use cases, making it an invaluable technique tool in computational geometry. To help you find an optimal library for mesh simplification tasks (in Python, C++, or both), our team conducted a practical search and will share some findings below.

Two Facets of Mesh Simplification

Mesh simplification is, of course, a very general term. That is to say, its practical extent varies, depending on the application. In this piece, we are going explore the following two key directions:

Heavy mesh simplification. Here, we reduce meshes by 1000x or more (e.g., 2M to 2K triangles). This approach is typically resorted to in scenarios where speed and low resource usage are critical;
Moderate mesh simplification, in turn, reduces meshes by 10x (e.g., 2M to 200K triangles). This balances performance with higher fidelity.

Challenges of Mesh Simplification

Heavy mesh simplification often implies significant risks, including holes, self-intersections, and degeneracies. Such structural issues could compromise visual fidelity, disrupt rendering workflows, and lead to functional problems in manufacturing or 3D printing;
Moderate simplification, while less extreme, presents its own pitfalls, i.e., maintaining geometric fidelity while reducing complexity. Striking the right balance between computational speed and model quality is critical to ensure the simplified surface actually remains usable.

Our Research Approach Explained

Different levels of mesh simplification are essential to match varying project needs. Understanding the trade-offs between speed and quality featured by numerous existing libraries is pivotal across both heavy and moderate cases.

We realise how difficult it is to compare libraries: you need to identify applicable ones, elaborate evaluation parameters, install, and run a testing scenario. FYI, our team took some popular libraries for comparison, from different standpoints, and contrasted them with each other for you:

Mesh Simplification Option	GitHub / Website	Primary Language(s)	Python Support	Notes
MeshLib (ST and MT)	Download on GitHub	C++	Yes	Provides a robust API for Python.
CGAL	Download on GitHub	C++	Yes	Python bindings available via SWIG.
LibIGL	Download on GitHub	C++	Yes	Offers Python bindings.
MeshLab	Download on GitHub	C++, JS	Yes	Python support through PyMeshLab.
Rhino 3D	Explore the website	C++	Yes	Supports Python via Rhino.Python.
Fusion 360 (Adaptive and Uniform)	Explore the website	C++	Yes	Unified API accessible from both Python and C++.
3DCoat	Explore the website	C++	Yes	Offers both a C++ Core API and a Python API.
Fast Quadric Mesh Simplification	Download on GitHub	C++	Yes	Designed for C++; Python bindings can be created through wrapping techniques.
MeshOptimizer	Download on GitHub	C++	Yes	Provides Python bindings.
VTK	Download on GitHub	C++	Yes	Offers Python wrappers for its functionalities.

As for our object to run a test on, we opted for the iconic Nefertiti mesh (2M triangles) as the benchmark notable for its complexity and relevance in 3D modeling flows.

All tests were conducted on a consistent hardware basis to ensure reliable comparisons:

Windows 11
Intel Core i7-12700H
32GB RAM
NVIDIA RTX 3060m (6GB VRAM)

Heavy Mesh Simplification Case

Again, for this test, our Nefertiti mesh (2 million triangles) was used. A reduction factor of 1000x was applied, resulting in a final mesh of 2000 triangles.

MeshLib ST

MeshLib MT

CGAL

LibIGL

MeshLab: Quadric Edge Collapse

MeshLab: Clustering

Rhino 3D

Fusion Uniform

3DCoat

Fast Quadric Mesh Simplification

MeshOptimizer

VTK

Moderate Mesh Simplification Case

For this test, our Nefertiti mesh (2 million triangles) was reduced by a factor of 10x, resulting in a final mesh of 200,000 triangles. This scenario evaluates how well libraries balance speed and quality while maintaining geometric fidelity in a less extreme reduction scenario.