intersects_id can return duplicate hits when mesh.ray._scale is very small

[hoyeong-heo]

Bug: Duplicate intersections due to too-small ray offset (self._scale)

I encountered an issue where trimesh.ray.intersects_id returns massive duplicate intersection points when multiple_hits=True.

Symptoms

• A single ray produces a very large number of duplicate hit locations.
• Increasing max_hits results in proportionally more identical points.
• All returned intersection points are effectively the same (within floating-point precision).

Root cause (suspected)

The problem appears to be caused by an extremely small self._scale value, which makes the computed ray offset:

distance = np.clip(_ray_offset_factor * self._scale, _ray_offset_floor, np.inf)

too small relative to the scene scale.

As a result:

• The updated ray origin (new_origins + ray_offsets) does not move far enough.
• The next Embree query hits the same triangle again.
• This repeats until max_hits is reached, producing duplicate intersections.

In short: ray origins are not effectively advancing between iterations.

Expected behavior

Ray origins should advance enough to avoid immediately re-hitting the same triangle (or duplicates should be filtered).

Environment

• trimesh version: 4.11.2
• Backend: Embree
• OS: (fill in)

Reproduction

I will provide:

• Minimal test script

import trimesh


mesh = trimesh.load_mesh("combined.stl")

x = -164.70464393956098
y = 275.5539451575655
z = 10000

origin = [[x, y, z]]
direction = [[0, 0, -1]]

# mesh.ray._scale_to_box = False

index_tri, index_ray, locations = mesh.ray.intersects_id(
    origin, direction, multiple_hits=True, return_locations=True
)

print(f"{index_tri=}")
print(f"{index_ray=}")
print(f"{locations=}")

• combined.stl mesh that reproduces the issue

combined.zip

Notes

This seems related to scale-dependent robustness in the multi-hit ray loop.

Potential fixes might include:

• A configurable option for minimum offset
• Triangle ID re-hit prevention