points-lsd

Point-seeded LSD (Line Segment Detector) bindings used by UPAL (Unified and Efficient Point-Line Local Features, ECCV 2026).

This is a fork of pytlsd by Iago Suárez, itself a pybind11 wrapper around the original C implementation of LSD by Rafael Grompone von Gioi. On top of the upstream bindings it adds lsd_from_points, which grows LSD regions only from a set of seed pixels (e.g. learned keypoints) instead of scanning the full image, and accepts externally computed gradients so that a learned line field can steer the detector.

Install

Prebuilt wheels are published for Linux (x86_64, aarch64), macOS (arm64, x86_64) and Windows (AMD64), CPython 3.10–3.14:

pip install points-lsd

Building from source needs a C++17 compiler and CMake ≥ 3.15 (no other native dependencies):

git clone --recursive https://github.com/francois141/points_lsd.git
pip install ./points_lsd

Usage

lsd_from_points needs the image, integer (x, y) seed pixels, and gradient norm/angle maps. The maps are float64 arrays of the image's shape with undefined pixels set to -1024.0; the snippet below computes them the way LSD does (2x2 finite differences), but any gradient, such as one derived from a learned line distance field, can be supplied.

import numpy as np
import points_lsd

gray = ...  # H x W float64 (or uint8) image
seeds = np.array([[x0, y0], [x1, y1]], dtype=np.int32)  # (x, y) pixel seeds inside the image


def lsd_gradients(img, not_defined=-1024.0, threshold=5.2262518595055063):
    norm = np.full(img.shape, not_defined)
    angle = np.full(img.shape, not_defined)
    a, b, c, d = img[:-1, :-1], img[:-1, 1:], img[1:, :-1], img[1:, 1:]
    gx, gy = b + d - a - c, c + d - a - b
    norm[:-1, :-1] = 0.5 * np.hypot(gx, gy)
    angle[:-1, :-1] = np.arctan2(gx, -gy)
    angle[norm <= threshold] = not_defined
    return norm, angle


gradnorm, gradangle = lsd_gradients(gray.astype(np.float64))

# N x 5 float32 array: [x1, y1, x2, y2, p] per segment, where p is LSD's angle precision.
segments = points_lsd.lsd_from_points(gray, seeds, 1.0, 0.6, 0.0, gradnorm, gradangle)

# The upstream full-image detector is still available (gradients optional there).
segments = points_lsd.lsd(gray)

Seeds outside the image, seed arrays that are not N x 2, or missing gradient maps raise ValueError / TypeError. Region growing starts only from seeds whose gradient angle is defined, so seeds should lie on (or one pixel before) an intensity edge.

Development

pip install -e ".[test]"
pytest tests            # test_smoke.py is numpy-only; test_lsd.py needs the [test] extra

Wheels and publishing

Supported wheels

Built with cibuildwheel (config in pyproject.toml under [tool.cibuildwheel]) for CPython 3.10, 3.11, 3.12, 3.13 and 3.14:

Platform	Architectures	Wheel tag
Linux (glibc ≥ 2.24)	x86_64, aarch64	`manylinux_2_24_.manylinux_2_28_`
macOS	arm64 (≥ 11.0), x86_64 (≥ 10.9–10.15, per Python build)	`macosx_*`
Windows	AMD64	`win_amd64`

That is 25 wheels plus an sdist per release. Not built: musllinux (Alpine), PyPy, free-threaded CPython and 32-bit platforms. On a compatible platform, pip can build from the sdist with a C++17 compiler and CMake ≥ 3.15. The extension has no required external native libraries. OpenMP support and the OpenCV-dependent C++ tests are opt-in and disabled for wheels.

How a release works

.github/workflows/wheels.yml runs on every push to main, every pull request, every v* tag, and on manual dispatch. Every run builds all wheels and the sdist, smoke-tests the wheels in clean environments where the runner can execute them (macOS x86_64 is skipped on the arm64 runner), and runs the full test suite on Linux, macOS and Windows. What happens with the artifacts depends on the trigger:

Trigger	Publishes to
push to `main` / pull request	nothing (build + test only)
manual run (Actions → Wheels → Run workflow) with Upload to TestPyPI ticked	TestPyPI
push of a tag `vX.Y.Z`	PyPI

Uploads use PyPI trusted publishing (OIDC), so no API tokens are stored. Before the first upload to an index, configure that index to trust this repository's wheels.yml workflow and its corresponding pypi or testpypi GitHub environment.

The version comes from pyproject.toml, not from git. The tag only triggers the upload, and the check_version job refuses a tag that does not match project.version (tag v0.1.0 ↔ version = "0.1.0"). A version can be uploaded to an index more than once only with new distribution filenames. Existing files cannot be overwritten, so replacing published artifacts requires a new version number.

Release checklist:

# 1. update project.version in pyproject.toml and land it on main
git add pyproject.toml
git commit -m "release: 0.2.0"
git push origin main                                 # CI must be green

# 2. optional dry run: Actions > Wheels > Run workflow > "Upload to TestPyPI", then
pip install -i https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple "points-lsd==0.2.0"

# 3. publish
git tag v0.2.0
git push origin v0.2.0                               # publish_pypi job uploads to PyPI

Building wheels locally

pipx run cibuildwheel --platform linux    # needs Docker; native arch, or set CIBW_ARCHS_LINUX
pipx run cibuildwheel --platform macos    # needs the python.org CPython installers
CIBW_BUILD="cp313-*" pipx run cibuildwheel --platform linux   # one interpreter only

cibuildwheel's Windows build must run on Windows (or in CI). A plain pip wheel . builds a single wheel for the current interpreter without cibuildwheel's portability repair and compatibility checks.

License

The binding code and the lsd_from_points extension retain the MIT license of upstream pytlsd (see LICENSE), while pybind11 is BSD-3-Clause. The bundled core detector src/lsd.cpp is © Rafael Grompone von Gioi and licensed under the GNU Affero General Public License v3 or later (see LICENSES/AGPL-3.0-or-later.txt); it is linked statically, so the distributed wheels as a whole are subject to the AGPL terms.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

points_lsd-0.1.0.tar.gz (327.5 kB view details)

Uploaded Aug 28, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

points_lsd-0.1.0-cp314-cp314-win_amd64.whl (328.3 kB view details)

Uploaded Aug 28, 2026 CPython 3.14Windows x86-64

points_lsd-0.1.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (133.6 kB view details)

Uploaded Aug 28, 2026 CPython 3.14manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

points_lsd-0.1.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (121.1 kB view details)

Uploaded Aug 28, 2026 CPython 3.14manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

points_lsd-0.1.0-cp314-cp314-macosx_11_0_arm64.whl (111.2 kB view details)

Uploaded Aug 28, 2026 CPython 3.14macOS 11.0+ ARM64

points_lsd-0.1.0-cp314-cp314-macosx_10_15_x86_64.whl (123.7 kB view details)

Uploaded Aug 28, 2026 CPython 3.14macOS 10.15+ x86-64

points_lsd-0.1.0-cp313-cp313-win_amd64.whl (318.7 kB view details)

Uploaded Aug 28, 2026 CPython 3.13Windows x86-64

points_lsd-0.1.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (133.6 kB view details)

Uploaded Aug 28, 2026 CPython 3.13manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

points_lsd-0.1.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (120.6 kB view details)

Uploaded Aug 28, 2026 CPython 3.13manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

points_lsd-0.1.0-cp313-cp313-macosx_11_0_arm64.whl (111.2 kB view details)

Uploaded Aug 28, 2026 CPython 3.13macOS 11.0+ ARM64

points_lsd-0.1.0-cp313-cp313-macosx_10_13_x86_64.whl (123.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.13macOS 10.13+ x86-64

points_lsd-0.1.0-cp312-cp312-win_amd64.whl (318.6 kB view details)

Uploaded Aug 28, 2026 CPython 3.12Windows x86-64

points_lsd-0.1.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (133.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.12manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

points_lsd-0.1.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (120.6 kB view details)

Uploaded Aug 28, 2026 CPython 3.12manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

points_lsd-0.1.0-cp312-cp312-macosx_11_0_arm64.whl (111.1 kB view details)

Uploaded Aug 28, 2026 CPython 3.12macOS 11.0+ ARM64

points_lsd-0.1.0-cp312-cp312-macosx_10_13_x86_64.whl (123.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.12macOS 10.13+ x86-64

points_lsd-0.1.0-cp311-cp311-win_amd64.whl (315.9 kB view details)

Uploaded Aug 28, 2026 CPython 3.11Windows x86-64

points_lsd-0.1.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (133.1 kB view details)

Uploaded Aug 28, 2026 CPython 3.11manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

points_lsd-0.1.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (119.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.11manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

points_lsd-0.1.0-cp311-cp311-macosx_11_0_arm64.whl (109.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.11macOS 11.0+ ARM64

points_lsd-0.1.0-cp311-cp311-macosx_10_9_x86_64.whl (121.1 kB view details)

Uploaded Aug 28, 2026 CPython 3.11macOS 10.9+ x86-64

points_lsd-0.1.0-cp310-cp310-win_amd64.whl (315.3 kB view details)

Uploaded Aug 28, 2026 CPython 3.10Windows x86-64

points_lsd-0.1.0-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (132.1 kB view details)

Uploaded Aug 28, 2026 CPython 3.10manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

points_lsd-0.1.0-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (118.9 kB view details)

Uploaded Aug 28, 2026 CPython 3.10manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

points_lsd-0.1.0-cp310-cp310-macosx_11_0_arm64.whl (108.2 kB view details)

Uploaded Aug 28, 2026 CPython 3.10macOS 11.0+ ARM64

points_lsd-0.1.0-cp310-cp310-macosx_10_9_x86_64.whl (119.5 kB view details)

Uploaded Aug 28, 2026 CPython 3.10macOS 10.9+ x86-64

File details

Details for the file points_lsd-0.1.0.tar.gz.

File metadata

Download URL: points_lsd-0.1.0.tar.gz
Upload date: Aug 28, 2026
Size: 327.5 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for points_lsd-0.1.0.tar.gz
Algorithm	Hash digest
SHA256	`4d71d7f077bbcd889768e883510208a43191066cb87bc02f7332942fa143c130`
MD5	`420e0d2950384e272e75b36ed2d00cbd`
BLAKE2b-256	`9d1aebd83c9f50de1e32e9074280ee46b310a44644700858207171d74f33754c`

Algorithm	Hash digest
SHA256	`4fd3e9958214614b5c498f0a8a45de2bb06a3f7e4d04043d85bc28b75cebba6f`
MD5	`69ba0e21b5f8aff44a25a53b33de17aa`
BLAKE2b-256	`72a4feee0101978fd0436849f0c5d2f98bf1f94fe07d44e1004092e2ffb9e689`

Algorithm	Hash digest
SHA256	`3477ec2a42fc4cf3bc9e3673042d3e77abbadc1db94606ae56c31542020cb56a`
MD5	`9285684d2e4ba5a4d9a31f8623afeb1c`
BLAKE2b-256	`d05c6c65fe362f904c8e32d34a602e09b238f949fbcebc6818dd8f325ba813b9`

Algorithm	Hash digest
SHA256	`bd78e136e74575a9e9c0f9cbc8b244d0c3bf7a7c789092bdcccab9f07ee80107`
MD5	`07002fd9491dcd19b4efc1e9323ebce0`
BLAKE2b-256	`4caed00bd43ed521dddf757512b2f1a3dde14cded23d4a0d717209bcd5b4a1f2`

Algorithm	Hash digest
SHA256	`1087fc7d18d75ac02b6b80f84274ae5109977dd8fc7a6d29e4bec006242f2209`
MD5	`8797f436ddf3092b2545c4a27689aa27`
BLAKE2b-256	`b5f9fa7c5d92912d92ce30770918fc70d8a4ace88680271015f9cca1219b1c90`

Algorithm	Hash digest
SHA256	`2e7dd1226d42645b2081af858fc05e137fd1f21455f23aef58d80b2d74875e3f`
MD5	`a48b54f1937115d5955f868fa7980b4c`
BLAKE2b-256	`5d1e9a74ebb20bd76a627c4361cae537f40c3968ca0b7d92f24ea57e2d432765`

Algorithm	Hash digest
SHA256	`0b004a659263b79f576996cd45629e29379d775b8f2c247bd6f0435b9373db36`
MD5	`ac8dc9a93ce310d0d6e750cb3752db9c`
BLAKE2b-256	`f49b095bac39b610157ad8b5c0946093c22e1abf42cf18ad16f7779fbd1a21cd`

Algorithm	Hash digest
SHA256	`12108a74f2749e4e77843322d5381ca10b5a22a29fa27952bcf9d4b9ac3db40b`
MD5	`ee88183cf98ecf493603d713248edff8`
BLAKE2b-256	`c62036e85c9d75f01307bb24fdc653e7410f90bfd3c2c35d1a6448e9b90f0a37`

Algorithm	Hash digest
SHA256	`6dce4f0675b36930be4536f9a356f22235c8944cf9733b4423b73f3e8d5748c3`
MD5	`fec9f6c207fb75cf82427127db25dfe9`
BLAKE2b-256	`779eec8cbb4d2b833600e90c55917413ac4e37f89ef5ebc73c5d41871cf7c5ba`

Algorithm	Hash digest
SHA256	`9c90b6e937b1c98737be5417ed8b00e6490a5ccf74a067af612774843ba33fe2`
MD5	`3fd61ea0e00fddd80cc1126ec5a732b9`
BLAKE2b-256	`f1949aa94581229bffcf4a79432f6d6bf2f24f01c9b4a9d1f89b20f1a0d44f81`

Algorithm	Hash digest
SHA256	`099deaa0a60f6eac02220ed4039b9594f1363dfa02531c17df1b8393285cff08`
MD5	`e895d9c1bf7d7f25a2328fff8bcba3f2`
BLAKE2b-256	`33985a8019fb4813541cdf0f9f1e8d0238675c654b45374e2f4f90f8ff88dd1f`

Algorithm	Hash digest
SHA256	`cb5321cbafe5a4cca2c0b199fd69e6d887d3b59b18717ed7c1204b4685601f19`
MD5	`1313b88c3461608a88fcf65910b2234c`
BLAKE2b-256	`b614bcf19c5cf4861b70280ed3061ddf3e0e1dd673db46727aad7cf21e9278dd`

Algorithm	Hash digest
SHA256	`4e4eec8c0189a96820da020cde38b58536b447974333043abb94a86e11d12ff5`
MD5	`664d8e9ca0a6fe6e6bd95228a6465f44`
BLAKE2b-256	`9a0362c2eca475769b081a7bdc7e89530dc9b63c346a6cf9249c3f3d25b1b7cb`

Algorithm	Hash digest
SHA256	`6b875441b65657fb73187ab791a1bea876fd8df6fb8a776f7bca0d8f1e057aeb`
MD5	`d600063ac3b4bc7e1b1dd43b6c12ec8e`
BLAKE2b-256	`be4618b09f7daa45b840991d2f017a97ee29ce3e3ca1f3eadf5c411a83ab02ab`

Algorithm	Hash digest
SHA256	`0f8f844539d387ecbe3ada1de7e88f98c52fe01bd0e67d8edb77fb25fe879cd8`
MD5	`1aa3e588dfa16c196b17ef260220b77f`
BLAKE2b-256	`c7ce1f6ecc61692895911b1eadf3439fa9507c8b69fa0efca17d93a5cc68ed57`

Algorithm	Hash digest
SHA256	`5575ec68e787f254a671e237caf02975ed67a9db63010f57ab14bf59af51f896`
MD5	`583e2b31c4a7a29d8018cc7cb81b08bc`
BLAKE2b-256	`3742d4943348346c5ae669120c370f70bc84083cfa7d2539870a0f7c5a0c16e2`

Algorithm	Hash digest
SHA256	`ee16ed059478eaca9edf266947d02ae67422de0dbd9332832c22ff5a6de66672`
MD5	`827d37ee21f166af3a45752408dbc0e1`
BLAKE2b-256	`960d5c0057eb48cf1c895e272bb046d34152f0b8a2522fddfa381adfbeeca4af`

Algorithm	Hash digest
SHA256	`7468df54011567597229105616df9d855e695d6bcf12080373e49eef6d12120c`
MD5	`3fdcac02149baf1aa996019e68515d3c`
BLAKE2b-256	`3f2b23b251a5037290d8edfd4116c44d5ede2deef8d6f37c59bc0ea4da9a80e0`

Algorithm	Hash digest
SHA256	`23af46d44e65fbb60b920520ff637dc8356410d8a4226216f42b61988b9ed6be`
MD5	`98189e30c1f535d913716ed17723e85c`
BLAKE2b-256	`4253b2f0dfade2eb867e9fd4b66d1bb185e0cdd1d5c8a804a386e0833c7fa072`

Algorithm	Hash digest
SHA256	`2c3ad884b212d05936a89f1542fc66002dcfa59ab809ab0d6f34e17dfbd3917d`
MD5	`c377beb8dba4a27e2bb1840c308ce0b8`
BLAKE2b-256	`32bf53dad3cef6147aed73413933eae88b1f27a40e156da5143a1c94393e40d9`

Algorithm	Hash digest
SHA256	`444556c5a441a1dd613acd90bed8a863447ffbf756e4b33c1af6e35848a28f80`
MD5	`e2d0bdcf0eb1c5ce8dd7d783c49d605a`
BLAKE2b-256	`8d95f0754aa2af70949e14f350e954b3f2542e670d5382070a8feb67bd29ef76`

points-lsd 0.1.0

points-lsd

Install

Usage

Development

Wheels and publishing

Supported wheels

How a release works

Building wheels locally

License

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License expression

Requires

Provides Extra

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

Algorithm	Hash digest
SHA256	`a09266c2e47de7302df3ab92d67814eeff9b94a047c69223bf227eda5036544d`
MD5	`905f276116c7f4bc75ab9e93ec3fd31b`
BLAKE2b-256	`fc38959ee29b6bcc497c372d2f7df584025c08b3a3e5d14680809b2efa10cd7e`

Algorithm	Hash digest
SHA256	`ecb543aa3171a4bb769f1bffb3eed7f8b1e71cedb00da604805e07163dc2f4bf`
MD5	`c86046e1647961af9faf1165a0f49183`
BLAKE2b-256	`155f647dfd0d3588833734a0c3de0536f0caf51f3051ffe4e9feb1787c09aa6d`

Algorithm	Hash digest
SHA256	`2e0855e27f33a697e43c05316166eee5a7c87e11407d1e1c8ce164b6c55fb581`
MD5	`7d1b02c4632868b34bd638a3790d5279`
BLAKE2b-256	`d74a1c7b8b3fd6b868f9a3881151877a42accf6db830016ac558b69e2eedcf52`

Algorithm	Hash digest
SHA256	`e2febd6ee1a6a8bf4ecf934488d2eb42a3e1d5c91035a7b1dec0bb1a52a8cc34`
MD5	`b5323b8a1e82f7b55e5f08aa75989137`
BLAKE2b-256	`6445f48d42e52e2920944342ea1db9374f021f52fcf3c763ed9cd6f2adb6d879`

Algorithm	Hash digest
SHA256	`86dd8298aafadec78d8bd1b35c5ce20b495d502cd6db0e0a9e9058ea801d3be8`
MD5	`484a44c10225c0c2eee4b74ec53bd7f2`
BLAKE2b-256	`11bcb98fdd78e01ee919c06aacce8efca8fa4b601e52f5fee4b178738643da4e`