Meshes
What’s a mesh?
A mesh is a collection of vertices, edges, and faces that define a 3D object.
This is how nearly all 3D is represented in real-world applications today.
The problem
Meshes are difficult for machine learning models.
Most 3D research today looks something like this:
- A machine learning pipeline produces a non-mesh 3D representation
- This non-mesh representation is converted to a mesh using Marching Cubes
While progress has been very rapid in Step 1, Step 2 is relatively unchanged since the 1980s. This creates a gap between 3D research, which produces advancements in Step 1, and 3D applications, which rely on the final output of Step 2.
This is a simplification, but it highlights the challenges in the 3D machine learning landscape broadly.
The solution
It remains to be seen whether Step 2 will be improved or if the 3D ecosystem will change entirely (or somewhere in between).
In the meantime, it’s useful to understand the building blocks of current 3D machine learning research, starting with non-meshes.
< > Update on GitHub