pyraymesh

Description

pyraymesh is a Python library for performing ray intersection and occlusion tests on 3D meshes using a Bounding Volume Hierarchy (BVH). The library uses the C++ library bvh for building the BVH and performing the intersection tests.

While this library is reasonably fast for simpler meshes (benchmarks coming soon), it is not as fast as Embree, espcially for larger and more complex meshes. However, it does not have any dependencies on external libraries, and is thus easier to install and use.

Installation

Install the package either by

pip install pyraymesh

or cloning the repo and using pip:

pip install .

Note that the package requires a C++ compiler to build the C++ extension.

Usage

Building the BVH

To build the BVH for a mesh:

from pyraymesh import Mesh
import numpy as np

vertices = np.array([[0, 0, 0], [1, 0, 0], [1, 1, 0], [0, 1, 0]])
faces = np.array([[0, 1, 2], [2, 3, 0]])
mesh = Mesh(vertices, faces)
mesh.build("medium")

The build method takes a string argument that specifies the BVH build type, which can be one of the following: "low", "medium" and "high". The build type determines the trade-off between build time and query time. For most cases "medium" is almost always the right choice.

Ray Intersection

To perform ray intersection tests:

ray_origin = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
ray_direction = [[0, 0, -1], [0, 0, 1]]
## or 
ray_origin = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
ray_direction = [0, 0, -1]  # multiple rays with same direction
## or 
ray_origin = [0.1, 0.2, 1]
ray_direction = [[0, 0, -1], [0, 0, 1]]  # multiple rays with same origin

result = mesh.intersect(ray_origin, ray_direction, tnear=0, tfar=1000)
print(result.num_hits)
print(result.coords)
print(result.tri_ids)
print(result.distances)

tnear and tfar can be scalars or lists of the same length as the number of rays. If they are scalars, the same value will be used for all rays. If they are lists, each value will be used for the corresponding ray.

If you set tnear to a value greater than 0, the intersection tests will ignore any intersections that are closer than tnear. Similarly, if you set tfar to a value less than infinity, the intersection tests will ignore any intersections that are farther than tfar. This library does not support negative values for tnear or tfar.

Reflections

If you want to get the reflection of the rays, add the calculate_reflections = True parameter to the intersect method:

ray_origin = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
ray_direction = [[0, 0, -1], [0, 0, 1]]
result = mesh.intersect(ray_origin, ray_direction, tnear=0, tfar=1000, calculate_reflections=True)
print(result.reflections)

results.reflections is a list of noramlized vectors representing the directions of the reflection of the rays. Only do this if you need the reflections, as it will slow down the intersection tests.

Occlusion Test

If you just care about whether a ray is occluded or not (i.e., you don't care about the intersection point) you can use the occlusion method which is faster than the intersect method and just returns an array of booleans.

ray_origin = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
ray_direction = [[0, 0, -1], [0, 0, 1]]
occluded = mesh.occlusion(ray_origin, ray_direction)
print(occluded)

Count intersections

If you want to know the total number of intersections for each ray along its path, without stopping at the first intersection, you can use the count_intersections method:

total_intersections = mesh.count_intersections(ray_origin, ray_direction)
print(total_intersections)

This method returns an array of integers representing the total number of triangles that each ray intersects between tnear and tfar.

Parallelization

The intersect and occlusion methods can be parallelized by passing threads parameter when calling the methods:

result = mesh.intersect(ray_origin, ray_direction, tnear=0, tfar=1000, threads=4)

The threads parameter specifies the number of threads to use for the intersection tests. If set to -1, the number of threads will be equal to the number of cores on the machine. In general you shouldn't set the number of threads to be greater than the number of cores on the machine.

For a small number of rays, the overhead of parallelization might make the parallel version slower than the serial version, so it is recommended to test the performance of both versions for your specific use case.

Ray Direction Utilities

The library includes several utility methods for generating ray directions distributed on a sphere. The sphere_direction_vectors function generates points evenly distributed on a sphere using a Fibonacci spiral pattern, providing excellent uniformity even with small sample counts. For an alternative distribution, hammersley_sphere_direction_vectors implements the low-discrepancy Hammersley sequence. When you need to sample within a specific angle, the cone_direction_vectors function creates rays distributed within a cone of a specified angle around a central direction. For completely random sampling, random_sphere_direction_vectors provides uniformly distributed random directions.

from pyraymesh.ray_functions import (
    sphere_direction_vectors,
    hammersley_sphere_direction_vectors,
    cone_direction_vectors,
    random_sphere_direction_vectors,
)

# Generate 1000 rays distributed on a sphere (using Fibonacci spiral pattern)   
sphere_rays = sphere_direction_vectors(1000) 
# Generate 1000 rays distributed on a sphere (using Hammersley sequence)
hammersley_rays = hammersley_sphere_direction_vectors(1000)
# Generate 1000 rays distributed within a cone of 30 degrees around the z-axis
cone_rays = cone_direction_vectors(1000, [0, 0, 1], 30)
# Generate 1000 uniformly distributed random rays
random_rays = random_sphere_direction_vectors(1000)

Test line-of-sight

If you want to know if two points are visible to each other, you can use the line_of_sight method:

origin_point = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
target_point = [[0, 0, -1], [0, 0, 1]]
## or 
origin_point = [[0.1, 0.2, 1], [0.2, 0.1, 1]]
target_point = [0, 0, -1]  # multiple origin points with same target
## or 
origin_point = [0.1, 0.2, 1]
target_point = [[0, 0, -1], [0, 0, 1]]  # multiple target points with same origin

visible = mesh.line_of_sight(origin_point, target_point)

visible is a list of booleans representing whether the target point is visible from the origin point.

Visibility Matrix

If you want to know the visibility matrix between all pairs of a list of points, you can use the visibility_matrix method: For N points it returns an NxN matrix where the element at (i, j) is True if the j-th point is visible from the i-th point.

points = [[0.1, 0.2, 1], [0.2, 0.1, 1], [0.3, 0.4, 1]]
vis_matrix = mesh.visibility_matrix(points)
# vis_matrix is a 3x3 array of booleans

Traverse the BVH

If you want to traverse the BVH and get all triangles that are along a ray in the BVH, you can use the traverse or traverse_all method. These are useful if you want to do some custom processing on the triangles that are potentially intersected by a ray. The traverse_all method returns a list of triangle IDs of all triangles potentially intersected by the ray. The traverse method returns a generator that you can use to traverse the BVH. If you know you will need all, or most, of the triangles, it is recommended to use traverse_all as it is faster. If you are likely to break early from the loop, you can use traverse fpr better performance and use less memory.

origin = [0, 0, 10]
direction = [0, 0, -1]

for t_id in mesh.traverse(origin, direction):
    print(f"Triangle {mesh.vertices[mesh.faces[t_id]]} is the first triangle in the BVH traversed by the ray.")
    break

all_triangles = mesh.traverse_all(origin, direction)   
for t_id in all_triangles:
    print(f"Triangle {mesh.vertices[mesh.faces[t_id]]} is potentially intersected by the ray.")

Testing

To run the tests:

pytest

License

This project is licensed under the MIT License - see the LICENSE file for details.