scanmountgoat / geometry_tools Goto Github PK

The current implementation of bounding sphere calculation works fine for meshes that have all welded vertices, but in calculation of meshes that have unwelded vertices, duplicate vertex positions offsets the bounding sphere center, resulting in the radius to be larger than is optimal.

I am presenting a solution to this, which is simple in concept: remove duplicate vertex positions from the positions list. The following Python script performs both the current calculations and new calculations and prints the result of both. The center is usually closer to vanilla, and at times, the most optimal, which can result in a more optimal bounding sphere radius.

import ssbh_data_py
import numpy as np

mesh = ssbh_data_py.mesh_data.read_mesh("model.numshb")

for o in mesh.objects:
    positions_mean_old = [0, 0, 0]
    positions_mean_new = [0, 0, 0]

    print(f'Name: {o.name}, Index: {o.sub_index}')

    # Current geometry_tools calculation
    for [x, y, z] in o.positions[0].data:
        positions_mean_old += np.array([x, y, z])
    print(f'Old: {positions_mean_old / len(o.positions[0].data)}')

    # Offered geometry_tools calculation
    for [x, y, z] in np.unique(o.positions[0].data, axis=0):
        positions_mean_new += np.array([x, y, z])
    print(f'New: {positions_mean_new / len(np.unique(o.positions[0].data, axis=0))}\n')

This script only generates the center of the sphere, but I'd have to bet the radius is more optimal as a result.

https://bheisler.github.io/criterion.rs/book/getting_started.html

Add the main functions to FFI and add documentation for usage and safety. There should be specific functions for different types like vec3, vec3a, etc since generics won't work across the FFI boundary.

The initial implementation can just be naive solution guaranteed to contain all the specified spheres.

This brings the features more in line with applications like Blender that allow users to customize the normals. The angle threshold is critical for hard surface modeling where an entirely welded mesh may still be expected to have sharp creases.

• investigate weighting by face area
• angle threshold (crease angle)
• specify certain edges as hard edges

Applications will want to use any of the primitive unsigned integer types depending on the vertex format. Signed integers are unusual and probably aren't worth supporting.

The current functions require creating an array or vector of glam::Vec3A, which likely isn't the format used for storage. This could be handled by adding conversion utilities and/or making the functions more generic.

Some users will want to use mikktspace for consistency between applications and game engines for normal maps. There isn't a safe Rust implementation at this time, and it doesn't make sense to develop and maintain one here. A possible solution is to link to another crate in the docs or wrap a crate with a similar API to the existing code in this repository.

This is a common format for some game data and should give similar performance to Vec3A.

• bounding spheres
• normals
• tangents