ultrafast-pycocotools

COCO evaluation in Rust, with a Python API compatible with pycocotools.

Use it to reduce evaluation time while keeping the reference metrics. The test suite compares the complete precision, recall, and scores arrays and the summary statistics byte for byte, including score ties, crowd annotations, custom evaluation parameters, and real COCO fixtures.

Benchmarks cover public pretrained detector outputs on COCO and a separate synthetic scalability workload on the public Objects365 dataset. Both compare complete evaluation arrays against pycocotools.

Status: 0.1.4, alpha. Validate your application's parameters and subclass behavior before replacing its reference evaluator.

Installation

Install from PyPI:

python -m pip install ultrafast-pycocotools

Prebuilt wheels cover CPython 3.9–3.14 on Linux x86-64 (glibc 2.17+), Windows x86-64, and macOS Intel/Apple Silicon. Compatible wheels require no Rust compiler. NumPy is installed automatically. Other platforms build from source.

The release workflow and maintainer guide describe wheel testing and Trusted Publishing.

Build from source

Requirements: Python 3.9+, a recent stable Rust toolchain, and a working native compiler toolchain. NumPy is installed as a dependency.

Clone the repository and build the package:

git clone https://github.com/developer0hye/ultrafast-pycocotools.git
cd ultrafast-pycocotools
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install .

On Windows PowerShell, activate with .\.venv\Scripts\Activate.ps1 instead. The source build uses Maturin and compiles the Rust extension; Rust is needed at build time, not when importing an already built wheel.

Reproduce without downloads

After installing the test/reference dependencies, run the complete synthetic check with one command:

python -m pip install ".[test]"
python bench/reproduce.py quick --out bench/out/quick --verify-published quick

It generates data, runs both scorers, and verifies input hashes and every evaluation-array byte. No images, model weights, or GPU are needed. See the reproduction guide for Objects365 and public detector predictions, expected hashes, resource requirements, and output files.

Quick start

This example uses the small COCO fixture included in the repository. Replace the two paths with your annotation and prediction JSON files for a real run.

from ultrafast_pycocotools import COCO, COCOeval

gt = COCO("tests/data/coco_subset_gt.json")
dt = gt.loadRes("tests/data/coco_subset_dt.json")
evaluator = COCOeval(gt, dt, "bbox")

evaluator.evaluate()
evaluator.accumulate()
evaluator.summarize()

print("AP:", evaluator.stats[0])

evaluator.run() is a convenience method for the last three calls. Standard COCO evaluation modes are "bbox", "segm", and "keypoints".

Faster loading and lower memory by default

Version 0.1.1 avoids allocating and retaining redundant polygons for box-only predictions. Ordinary gt.loadRes(predictions) gets both improvements; no performance flag is needed. Bbox/segmentation metrics and public mask/plot helpers remain covered by reference comparisons. Direct annotation dictionaries omit the derived segmentation field unless derive_segmentation=True is requested. Measurements against 0.1.0.

Version 0.1.2 further improves the COCO case by 8.9% in time and 1.9% in peak RSS relative to 0.1.1, using native index/metadata loops and releasing completed category buffers earlier. Repeated measurements and limits.

Version 0.1.3 adds compact bbox file loading: 2.6–2.9× faster and 56–62% less peak RSS than 0.1.2 on three tested workloads, including JSON parsing in both versions. Annotation dictionaries materialize only when accessed; evaluation shares immutable bbox coordinates. Existing dict/list inputs retain their previous representation and show no demonstrated speed/memory improvement. Measurements, API behavior and lower bounds.

YOLO26n file evaluation time and peak memory versus pycocotools and faster-coco-eval

Version 0.1.4 further reduces Rust allocations with borrowed index/coordinate slices, smaller detection records and a smaller recall workspace. Repeated measurements and implementation details.

LVIS and metric names

gt = COCO("lvis_val.json")
dt = gt.loadRes("predictions.json")
evaluator = COCOeval(gt, dt, "bbox", lvis_style=True)  # also supports "segm"
evaluator.run()
print(evaluator.stats_as_dict["APr"])

LVIS uses its federated annotation protocol, global 300-detection image limit, and rare/common/frequent category metrics. Dictionary names follow the official LVIS API (AP, AP50, APr, AR@300, etc.); framework aliases such as AP_all and AP_50 are also available. Standard COCO stats positions are unchanged. LVIS verification and integration guide.

Use with an existing framework

Prefer direct imports when you own the evaluation code. If a framework imports pycocotools internally, register the replacement before importing that framework:

from ultrafast_pycocotools import init_as_pycocotools

init_as_pycocotools()

from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval

This changes the import mapping for the entire Python process. Use separate processes when comparing the reference package and the replacement.

Scaling with input size (0.1.0 measurements)

41.9× faster evaluation · 89.5% lower peak memory than pycocotools at the largest measured input (80,000 images). Also 8.1× faster with 92.2% lower peak memory than faster-coco-eval. These callouts use the recorded 0.1.0 runs below.

Evaluation time and peak memory versus GT plus prediction count

Measured on nested Objects365 subsets with identical inputs for both scorers. Ultrafast matches pycocotools byte for byte at every point. Faster-coco-eval passes a separate numerical tolerance check; its arrays are not byte-identical. Method, counts and reproduction · SVG · PDF · Raw measurements.

RF-DETR integration

The optional adapter connects ultrafast to RF-DETR's actual one-pass training metric, which otherwise uses faster-coco-eval. Tests cover bbox/segmentation, per-class metrics, reset/pickle and two-rank CPU state merging. A separate RF-DETR Nano benchmark uses all 5,000 COCO validation images and 1.5 million real predictions. In the actual RF-DETR metric replay, it is 2.09× faster with 37.2% less peak RSS than the existing faster-coco-eval backend, with identical aggregate and per-class metric tensors. Measurements and opt-in integration.

RF-DETR metric replay speed and memory

YOLO26n benchmark (0.1.2)

19.1× faster · 53.4% lower peak RSS than pycocotools, using identical YOLO26n predictions on all 5,000 COCO val2017 images.

Scorer	Evaluation time	Peak RSS
pycocotools 2.0.11	37.073 s	1,599.2 MiB
faster-coco-eval 1.8.0	7.257 s	1,651.6 MiB
ultrafast-pycocotools 0.1.2	1.942 s	745.9 MiB

All four ultrafast evaluation arrays match pycocotools byte for byte. The additional backend is within absolute 1e-12 tolerance but not byte-identical. The 596,202 predictions were generated on CPU in about four minutes; inference is excluded from evaluation times. One fresh process per scorer on a shared host, two CPU cores each. Settings, hashes and reproduction.

Public benchmarks (0.1.0 measurements)

The same saved inputs are scored by pycocotools 2.0.11, faster-coco-eval 1.8.0 and ultrafast-pycocotools 0.1.0. Ultrafast matches the reference arrays byte for byte. Faster-coco-eval agrees within absolute tolerance 1e-12 (rtol=0), with small floating-point differences reported explicitly.

Workload	pycocotools	faster-coco-eval	ultrafast	Ultrafast speedup vs faster-coco-eval
COCO val2017 / public YOLO11m	29.87 s	4.88 s	2.61 s	1.87×
Objects365 v2 / synthetic predictions	520.26 s	100.51 s	12.43 s	8.09×

Workload	pycocotools RSS	faster-coco-eval RSS	ultrafast RSS
COCO val2017 / public YOLO11m	1.28 GiB	1.34 GiB	0.70 GiB
Objects365 v2 / synthetic predictions	22.62 GiB	30.46 GiB	2.37 GiB

Measured with Python 3.12.3 and NumPy 2.4.4 on an AMD EPYC 9554 host, with two CPU cores available to each scorer, two Rayon/OpenMP threads and one OpenBLAS thread. One run per case; prior reference/ultrafast results are reused and the additional backend is measured afterward. These are shared-host observations.

Time includes GT indexing, result loading, evaluation, accumulation and summarization. It excludes JSON parsing, inference and output serialization. Memory is whole-process peak RSS, including parsed inputs and serialization. Objects365 predictions are synthetic: this is evaluator scalability, not trained detector accuracy or end-to-end Ultralytics validation speed.

Three-backend comparison and reproduction · Raw results and hashes · General reproduction guide.

Compatibility and intentional differences

The public COCO, COCOeval, and mask APIs are checked against pycocotools. Compatibility tests cover query ordering, result loading, RLE formats, evaluation parameters, and subclass overrides. Full-array comparison matters: rounded AP can hide differences in individual precision cells.

Some implementation details intentionally differ:

evaluate() performs matching and accumulation internally; accumulate() exposes the result through the standard API.
Per-image evalImgs records are not stored by default. Use COCOeval(gt, dt, "bbox", store_eval_imgs=True) if your integration reads them.
IoU and annotation lookup structures are built lazily when accessed.
Ground-truth annotation dictionaries are not rewritten in place.
COCO(path, verbose=False) suppresses loader progress output.
Box-only results omit redundant derived polygons by default; use derive_segmentation=True when reading that field directly.
COCO(annotation_dict) borrows the dictionary without a deep copy.

The evaluator follows pycocotools' treatment of iscrowd, including its handling of the annotation ignore field. Applications that rely on mutation side effects or unusual evaluation parameters should run their own parity checks. See the detailed compatibility notes.

Additional diagnostics

After evaluation, the same matching results support per-category statistics, precision–recall curves, match inspection, and a confusion matrix:

print(evaluator.stats_as_dict)
print(evaluator.per_category_stats())
curve = evaluator.pr_curve(cat_id=1, iou_thr=0.5)
matches = evaluator.matches(iou_thr=0.5)
matrix = evaluator.confusion_matrix()

Boundary IoU is available as an additional evaluation mode. It is an extension, not a standard pycocotools metric. See the implementation notes for custom thresholds, area ranges, and diagnostic output formats.

Development and verification

Write documentation, code comments, docstrings and examples in English.

Automated CI builds and tests Linux, macOS and Windows, checks NumPy 1/2 and multiple Python versions, and runs Rust tests in debug and release. The required CI status blocks main updates when any test job fails.

Install the test dependencies and run the reference-comparison suite:

python -m pip install -e ".[test,lvis-test]"
python bench/fetch_lvis_fixture.py
python -m pytest -q
cargo test -p ufcoco-core

The repository includes synthetic inputs and a small real COCO fixture. Tests requiring the complete dataset skip when those local files are absent; see fixture documentation. Exact parity claims refer to the tested inputs and configurations, not a proof over every possible input.

Use bench/compare_saved_predictions.py to save full evaluation arrays and compare your own predictions. For changes to matching or accumulation, retain byte-level parity tests rather than checking only the rounded summary.

License and credits

BSD-2-Clause. COCO evaluation algorithms and API compatibility are based on pycocotools, by Piotr Dollár and Tsung-Yi Lin, under BSD-2-Clause. Implementation design and numerical compatibility decisions are described in DESIGN.md. LVIS protocol verification uses the official LVIS API and its public example annotations/predictions; source revisions and hashes are recorded in the repository.

Download files

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

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

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.

ultrafast_pycocotools-0.1.4-cp314-cp314-win_amd64.whl (470.4 kB view details)

Uploaded Sep 8, 2026 CPython 3.14Windows x86-64

ultrafast_pycocotools-0.1.4-cp314-cp314-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (553.6 kB view details)

Uploaded Sep 8, 2026 CPython 3.14manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp314-cp314-macosx_11_0_arm64.whl (494.4 kB view details)

Uploaded Sep 8, 2026 CPython 3.14macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp314-cp314-macosx_10_12_x86_64.whl (521.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.14macOS 10.12+ x86-64

ultrafast_pycocotools-0.1.4-cp313-cp313-win_amd64.whl (471.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.13Windows x86-64

ultrafast_pycocotools-0.1.4-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (553.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.13manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp313-cp313-macosx_11_0_arm64.whl (494.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.13macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp313-cp313-macosx_10_12_x86_64.whl (521.4 kB view details)

Uploaded Sep 8, 2026 CPython 3.13macOS 10.12+ x86-64

ultrafast_pycocotools-0.1.4-cp312-cp312-win_amd64.whl (471.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.12Windows x86-64

ultrafast_pycocotools-0.1.4-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (553.6 kB view details)

Uploaded Sep 8, 2026 CPython 3.12manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp312-cp312-macosx_11_0_arm64.whl (494.9 kB view details)

Uploaded Sep 8, 2026 CPython 3.12macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp312-cp312-macosx_10_12_x86_64.whl (521.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.12macOS 10.12+ x86-64

ultrafast_pycocotools-0.1.4-cp311-cp311-win_amd64.whl (475.4 kB view details)

Uploaded Sep 8, 2026 CPython 3.11Windows x86-64

ultrafast_pycocotools-0.1.4-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (555.1 kB view details)

Uploaded Sep 8, 2026 CPython 3.11manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp311-cp311-macosx_11_0_arm64.whl (499.0 kB view details)

Uploaded Sep 8, 2026 CPython 3.11macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp311-cp311-macosx_10_12_x86_64.whl (513.6 kB view details)

Uploaded Sep 8, 2026 CPython 3.11macOS 10.12+ x86-64

ultrafast_pycocotools-0.1.4-cp310-cp310-win_amd64.whl (475.5 kB view details)

Uploaded Sep 8, 2026 CPython 3.10Windows x86-64

ultrafast_pycocotools-0.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (555.3 kB view details)

Uploaded Sep 8, 2026 CPython 3.10manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp310-cp310-macosx_11_0_arm64.whl (499.2 kB view details)

Uploaded Sep 8, 2026 CPython 3.10macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp310-cp310-macosx_10_12_x86_64.whl (513.9 kB view details)

Uploaded Sep 8, 2026 CPython 3.10macOS 10.12+ x86-64

ultrafast_pycocotools-0.1.4-cp39-cp39-win_amd64.whl (478.4 kB view details)

Uploaded Sep 8, 2026 CPython 3.9Windows x86-64

ultrafast_pycocotools-0.1.4-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (557.7 kB view details)

Uploaded Sep 8, 2026 CPython 3.9manylinux: glibc 2.17+ x86-64

ultrafast_pycocotools-0.1.4-cp39-cp39-macosx_11_0_arm64.whl (501.2 kB view details)

Uploaded Sep 8, 2026 CPython 3.9macOS 11.0+ ARM64

ultrafast_pycocotools-0.1.4-cp39-cp39-macosx_10_12_x86_64.whl (516.2 kB view details)

Uploaded Sep 8, 2026 CPython 3.9macOS 10.12+ x86-64

File details

Details for the file ultrafast_pycocotools-0.1.4-cp314-cp314-win_amd64.whl.

File metadata

Download URL: ultrafast_pycocotools-0.1.4-cp314-cp314-win_amd64.whl
Upload date: Sep 8, 2026
Size: 470.4 kB
Tags: CPython 3.14, Windows x86-64
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for ultrafast_pycocotools-0.1.4-cp314-cp314-win_amd64.whl
Algorithm	Hash digest
SHA256	`8f2a1546d9e2e9863b726d5fc6eba2be5dac0759f47d58438901a8e83b0d1e50`
MD5	`b4bcbaaf099c62674748552d329c54b5`
BLAKE2b-256	`c66be8d5a4e0d3a786941c415e9df8fd5716515a9b5907a78513c1ba3851b3e4`

Algorithm	Hash digest
SHA256	`41280b0d74ded89caba08ebede9d20824f77401725952f29c45f4ff563e3e7c4`
MD5	`266cf32352325b41a78116911b9f4cc3`
BLAKE2b-256	`93c4173eff0175d66fae4880fc3d5387173c5a7ed3953325f064c81443266e72`

Algorithm	Hash digest
SHA256	`ed9a112c5af16e7a072780ebe422e98ce37c9d28d0f4371257d6b8447599e28d`
MD5	`25abbd7880266107652fd752055a3fd9`
BLAKE2b-256	`11cb4fe96ab9ebfdd1de3d768a38af95a92278f426047dd6a87785ce090857a8`

Algorithm	Hash digest
SHA256	`e345930eea83d5f1388ce14bf42717810a5571ab6108de7a5724daee8d725616`
MD5	`340f90a9cccd1c4c61549f24c9440052`
BLAKE2b-256	`c18610abec7bbf783806489fa300cb9a3b351dff4281022a68711d17928a4cde`

Algorithm	Hash digest
SHA256	`d1d3d71212f505ff2a3d8f415f1aaacac8b7c8cec6f05e03c728d3df59954487`
MD5	`30b11542fa973f5364d154aab818fd5d`
BLAKE2b-256	`08417e86e0b8089e2735d112c8a76e76c34ad7df9e4a4a78a255022002931ff2`

Algorithm	Hash digest
SHA256	`924b5bdfeda7cec994c347302fc8972ccbb491d5bb619be3d83c83369f74d178`
MD5	`cee197b336faf7ae81fad382b06a8bc5`
BLAKE2b-256	`824b42a59cce89cad69214507286c70d2bf3ad12947ec546e7197727d152f916`

Algorithm	Hash digest
SHA256	`6bede5105019c7eca19088bab9f7c8e046b71a2abcab4fadcc65c30b58b63c5e`
MD5	`c6860ca11f74b28894172d4f9197c621`
BLAKE2b-256	`11d15dfdc337633be8cc7f0b890cf7a9eda6f287cbfb8c57d41c6c595cdd7935`

Algorithm	Hash digest
SHA256	`b465b090a6a7ce021f48438a2c5d9613e7a63796e75b812be3f2c57c18a05e3c`
MD5	`8e429e8822a0f11c076104fab2eb3611`
BLAKE2b-256	`6d3f40b3841b5c8c9b5dc135e55ee0878878ec326a5495525525acb3b3ae399d`

Algorithm	Hash digest
SHA256	`619999d7608218fc2b26af8a5ae6b98f0803aaa1f7a9d7e25fb37974bc67f13a`
MD5	`f70194070c01d0b842508abd6f20931a`
BLAKE2b-256	`2e1f20fbc35c915ca3514edf2f96edb8a1f787aacdc1c5b8b4c393a6b3060064`

Algorithm	Hash digest
SHA256	`58cf84f03f607ecd2594c8bf100a421483184d180500181c43d0efb5bb2810ec`
MD5	`6cd6ae7fbbd6d89a2439aeb45e50c6da`
BLAKE2b-256	`600221ddf29b16d116505adbea7122f5b3a27f87e967e45a11f4195df3b7a26a`

Algorithm	Hash digest
SHA256	`ce5d93eb3fd2ba79e31d621c92dfeb000ebc994f0dd97722d0ab4034cab687e6`
MD5	`9a02730d6169f333a1f17e5bdce55f08`
BLAKE2b-256	`56fb103550ec3afed429a4adaf459518af40d8af6868b06173b112de42cc071b`

Algorithm	Hash digest
SHA256	`2f791ab13bd08d2f62b2a082510278a5cd3ad5b91766f306dcb9d7a8b01f26b9`
MD5	`c41355a2e3ff5a2d872ffa5ababb3c71`
BLAKE2b-256	`96ff69ddf9503dd2f45806ad9922b82539a71a360daf7945eb0108f91a3b7d9f`

Algorithm	Hash digest
SHA256	`55bd93b87935482ffb5aa7f7a4ad51baa4be9bf7ead977ff03370cb5e4825575`
MD5	`71ba52bb929fbae2a9710af8f94ef78f`
BLAKE2b-256	`a734c388be3a35835982389998008ea9795108003e7ebbf9820036b16707a540`

Algorithm	Hash digest
SHA256	`3560952ead9683de4df99cd1bb7f2ce8c102325db66f3f99152f665f28ad3e6f`
MD5	`92091274de54f227e6c8b7498d1a09e6`
BLAKE2b-256	`d3dd3cd52a28d1140a439558fa2e8ec4ef2e06ab22fbd3476c9e6f097bf31de8`

Algorithm	Hash digest
SHA256	`3f38252dc4d6090e87a46c09320f1a2dd661a2cd40ed28f056e6a8c7e2973197`
MD5	`aac02e9568de3564d841eeeee43988e2`
BLAKE2b-256	`66050385a4af4ae5c649fcceb60d3593f1c9ba013a50d5d1675e7f38b57daf48`

Algorithm	Hash digest
SHA256	`749a13c0974d3db6d6697cb675448a288662ca5357f5d3cb8340e9f444c23fb9`
MD5	`9d46cf0e5704520343205eb353bdc502`
BLAKE2b-256	`a7195062c71eaf492edd160b832b7d5dfc1ba2724473a1391f51476cacc084ba`

Algorithm	Hash digest
SHA256	`f1167808496bbd1e04fe51967bc8c9374d342bfc98f052de578b30622880c172`
MD5	`8ef566a3ec3ccda2edc90d02705afc51`
BLAKE2b-256	`f68a3551e978774d6b6380c683e4e381fdf813f0b4b4b9e89c2c622bf3c9523c`

Algorithm	Hash digest
SHA256	`430482d448214afa55180dc274fe75a17087866460886810406b20a01ef523ed`
MD5	`5bce2bbbd461d4866f3e4ccdee5b2096`
BLAKE2b-256	`f0524cc4487b4590ff0ee98f804c452eb83ce54559b87f988467dae1e0f0632f`

Algorithm	Hash digest
SHA256	`d12e5841c4fdf681c40842880b5a3a0c5b7d50c5d9404aa0ed725080bc3f8b25`
MD5	`68b1a1265ee80681acc3d72017f15cef`
BLAKE2b-256	`24efcc9a096e2878927f8f6434ade44f5610b5b953a8516c54691184476be63f`

Algorithm	Hash digest
SHA256	`e4d1a38f0d464963047ec432879e1c717badf8f632238d89b0df980b90780418`
MD5	`d45848a6e18b5b8031673f2bb8cff51a`
BLAKE2b-256	`8db6723f74ca34ba50ae974b1666aee7bf9ea81564e4ab7b8e7f9c54107b80f6`

Algorithm	Hash digest
SHA256	`6cba6af1860cd11d6c9521420acd9bd0a6f24209d193fc2573dbefa138bada77`
MD5	`161bd3b24b8448f088fd8c52d9f467d0`
BLAKE2b-256	`75262d30d91f8c794fb2cb938cc719be8ff04707cee2b30a4caf0803d379b8d0`

ultrafast-pycocotools 0.1.4

ultrafast-pycocotools

Installation

Build from source

Reproduce without downloads

Quick start

Faster loading and lower memory by default

LVIS and metric names

Use with an existing framework

Scaling with input size (0.1.0 measurements)

RF-DETR integration

YOLO26n benchmark (0.1.2)

Public benchmarks (0.1.0 measurements)

Compatibility and intentional differences

Additional diagnostics

Development and verification

License and credits

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License

Requires

Provides Extra

Tags

Classifiers

Download files

Source Distributions

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

Algorithm	Hash digest
SHA256	`59a4977ce6d59fb12785f76e804c2d19ef6bd815f42d42f9d22408156b2012ae`
MD5	`091ec4ac5b7a0e93ec3822a7d49faeb6`
BLAKE2b-256	`d90a7c76dee8b3f06fbbed606011496e3aad0e47235947f2c201a8b827c48c1c`

Algorithm	Hash digest
SHA256	`d12e96ce653012c086106603392de52702cf6411721fc7db72cf0fa267ec7e4a`
MD5	`eb0367daf1c92d0d6d96491d7b98df3d`
BLAKE2b-256	`f1cd17dc0dbc9cb20daa757102499fe8e8e65ba698e24056c47452440170acdd`

Algorithm	Hash digest
SHA256	`1eae0ee6d48206a97167bd0f8fd5e4829bddeb255d9e77ca1fb4fbae4f9ef993`
MD5	`f959837adaf367d3993f5431720301e8`
BLAKE2b-256	`86095ce0525ebb8b5dd5511cc8c13be7611678e9fb2529bba1306cf99f6ef341`