Fix: Improve performance for non-contiguous NumPy arrays in pybind vector converters

[Khizar1]

…rs_*

Type

• Bug fix (non-breaking change which fixes an issue): Fixes open3d.utility.Vector3dVector() sometimes slow #5250
• New feature (non-breaking change which adds functionality). Resolves #
• Breaking change (fix or feature that would cause existing functionality to not work as expected) Resolves #

Motivation and Context

This PR fixes slow performance when constructing Open3D vectors (Vector3dVector) from non-contiguous NumPy arrays, such as Pandas DataFrames. Previously, pybind11 performed slow element-by-element deep copies due to the restrictive py::array::c_style flag in py_array_to_vectors_*. This change relaxes that restriction and performs a lightweight runtime check instead:

• If the array is already contiguous -> no copy.
• If not -> a single fast contiguous copy via NumPy.

Up to ~50× faster conversion for non-contiguous arrays. Please refer to benchmark results below.

Checklist:

• I have run python util/check_style.py --apply to apply Open3D code style
  to my code.
• This PR changes Open3D behavior or adds new functionality.
  • Both C++ (Doxygen) and Python (Sphinx / Google style) documentation is
    updated accordingly.
  • I have added or updated C++ and / or Python unit tests OR included test
    results (e.g. screenshots or numbers) here.
• I will follow up and update the code if CI fails.
• For fork PRs, I have selected Allow edits from maintainers.

Description

This PR improves array conversion performance in cpp/pybind/utility/eigen.cpp by replacing
the static py::array::c_style flag with a runtime check for contiguity.

Summary of Changes

• Updated py_array_to_vectors_double , py_array_to_vectors_int, py_array_to_vectors_int_eigen_allocator and py_array_to_vectors_int64_eigen_allocator to:
  • Detect C-contiguous arrays at runtime (array.flags()).
  • Copy only if the array is not c-contagious.
• No behavior change for contiguous inputs.

Benchmark Results

Vector Type	Before (ms, Pandas -> NumPy)	After (ms)	Speedup
Vector2dVector	2602.73	126.28	~21×
Vector2iVector	2502.89	49.43	~51×
Vector3dVector	2640.17	164.48	~16×
Vector3iVector	2607.42	69.98	~37×
Vector4iVector	2567.23	85.04	~30×

Reproducible script: benchmark_all_vectors.py

Tested on Ubuntu 24.04, Python 3.12 inside a Docker environment.

[update-docs]

Thanks for submitting this pull request! The maintainers of this repository would appreciate if you could update the CHANGELOG.md based on your changes.

[Copilot]

Pull Request Overview

This PR fixes a performance bottleneck when converting non-contiguous NumPy arrays (e.g., from Pandas DataFrames) to Open3D vector types. The fix removes restrictive compile-time flags and uses runtime contiguity checks instead, achieving up to 50× speedup for non-contiguous inputs while maintaining identical behavior for contiguous arrays.

Key Changes:

• Modified four conversion functions to perform runtime contiguity checks instead of compile-time enforcement
• Non-contiguous arrays now trigger a single fast NumPy copy instead of slow element-by-element deep copies
• Updated CHANGELOG to document the performance improvement

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
cpp/pybind/utility/eigen.cpp	Replaced static py::array::c_style flags with runtime contiguity checks in py_array_to_vectors_double, py_array_to_vectors_int, py_array_to_vectors_int_eigen_allocator, and py_array_to_vectors_int64_eigen_allocator
CHANGELOG.md	Added entry documenting the performance fix for non-contiguous NumPy array conversion

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

[nitpick] Add a space between '(issue #5250)' and '(PR #7343)' for consistency with formatting in other changelog entries.

Suggested change

	- Fix performance for non-contiguous NumPy array conversion in pybind vector converters. This change removes restrictive `py::array::c_style` flags and adds a runtime contiguity check, improving Pandas-to-Open3D conversion speed by up to ~50×. (issue #5250)(PR #7343).
	- Fix performance for non-contiguous NumPy array conversion in pybind vector converters. This change removes restrictive `py::array::c_style` flags and adds a runtime contiguity check, improving Pandas-to-Open3D conversion speed by up to ~50×. (issue #5250) (PR #7343).

[Khizar1]

@ssheorey I have fixed styling in eigen.cpp using python util/check_style.py --apply. This is the only file I have changed.
Some CI tests are still failing but appear to be unreleted to my changes. Let me know if any CI errors are caused by my changes.