NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #2944 most downloaded on PyPI
High-performance image processing functions for deep learning and computer vision.
Latest stable
0.3.5
Last release 17 days ago
05 Nov 2026
Release timing varies
gaps range from 8 days to 2 months
Nearly every release is documented
notes for 57 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
3 years old
67 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
Albucore 0.3.5 adds six NumPy image operations and fixes writable output from shared lookup-table calls.
Albucore 0.3.5 adds six NumPy image operations and fixes writable output from shared lookup-table calls.
box_blur, convolve, erode, dilate, linear_transformation_rgb, and unsharp_mask are now available from the package root. They preserve the input shape and dtype, return independent writable storage, and leave the source unchanged. Supported dtypes are uint8, float16, and float32; float16 inputs compute in float32 before restoring float16 storage.
Spatial filters accept HWC images with an explicit channel dimension, including (H, W, 1) grayscale. RGB matrix transforms accept HWC images or NHWC batches with one shared matrix. See the API contracts and PR #218.
apply_uint8_lut(image, lut) now returns independent writable storage when applying a shared (256,) table without out. Previously, this path could return a read-only array. Explicit-output and in-place calls retain their existing contract. PR #220.
Python and dependency minimums are unchanged from 0.3.4. Update an existing installation with its chosen OpenCV and PyTorch builds:
python -m pip install --upgrade albucore==0.3.5
Your coding agent can read these notes before it upgrades. Set up the MCP server →
This release reduces temporary memory in float16-to-float32 conversion and speeds up raw uint8-to-float32 widening on the measured Linux x86-64 path.
This release reduces temporary memory in float16-to-float32 conversion and speeds up raw uint8-to-float32
widening on the measured Linux x86-64 path. It also removes the public ImageType typing alias: applications
that import it must update their annotations before upgrading.
ImageType importsImageType is removed from both albucore and albucore._types. Replace it with NDArray annotations
that describe the dtypes your code accepts. For example, replace this float32 image annotation:
from albucore import ImageType
image: ImageType
with:
import numpy as np
from numpy.typing import NDArray
image: NDArray[np.float32]
For code accepting all three supported NumPy dtypes, use
NDArray[np.uint8] | NDArray[np.float16] | NDArray[np.float32]. Array shape and layout follow each operation's
contract; these annotations specify storage dtype. This change preserves the concrete dtype and container
relationships of public operation results.
Python and dependency minimums are unchanged from 0.3.3. To update an existing installation with its chosen OpenCV and PyTorch builds:
python -m pip install --upgrade albucore==0.3.4
The finiteness check used by the selected NumKong x86 cast path now reduces integer views of the input instead
of creating an image-sized mask. In the measured matrix, its largest temporary fell from 13.5 MiB to less
than 2 KiB. This saves working memory even when the caller reuses a float32 destination through out=.
An allocating conversion still needs storage for the float32 result. Conversion values, special-value
handling, and NumPy floating-point error behavior are preserved.
For contiguous NumPy input, convert_dtype(image, np.float32, max_value=1) now casts directly, avoiding lookup-table
construction and application. This is useful when a downstream operation expects floating-point pixel values
on the original 0–255 scale:
import numpy as np
from albucore import convert_dtype
pixels = np.array([[[0], [128], [255]]], dtype=np.uint8)
raw = convert_dtype(pixels, np.float32, max_value=1)
assert raw.tolist() == [[[0.0], [128.0], [255.0]]]
The default max_value=255 produces normalized floating-point intensities. Allocating calls preserve the source;
calls with out=destination return that destination. Other input layouts and scales retain their existing routes.
Representative public-call timings for 480 × 640 × 3 images on a 2.80 GHz Intel Xeon with AVX-512, using Python 3.11 and one native thread:
| Conversion | 0.3.3 | 0.3.4 |
|---|---|---|
| float16 → float32, allocate result | 0.478 ms | 0.379 ms |
float16 → float32, reuse out |
0.483 ms | 0.372 ms |
uint8 → float32, max_value=1, allocate result |
0.452 ms | 0.185 ms |
The half-cast rows use mixed positive and negative finite values. The complete matrix covers four image sizes from 128 × 160 to 768 × 1024, C=1/3/9, and allocating and explicit-output calls. Outputs match bitwise. Apple M4 Max public-call timings were essentially unchanged; no measured public cell on either platform slowed by more than 5%. See the complete benchmark and memory results for all cells, dependency versions, and reproduction commands.
Implementation and tests: PR #211 · Full changes since 0.3.3
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Area | Result | Evidence |
|---|---|---|
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Headless wheel smoke | pass | release-candidate workflow smoke test |
| Check | Result | Evidence |
|---|---|---|
| Router contracts | pass | tools/check_router_contracts.py |
| CI matrix policy | pass | tools/ci_matrix.py check |
| Golden vectors | pass | tools/verify_golden_vectors.py |
| Property tests | pass | pytest tests/property --hypothesis-profile=release |
| Artifact | Status |
|---|---|
| SBOM | present |
| SHA256SUMS.txt | present |
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Area | Result | Evidence |
|---|---|---|
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Headless wheel smoke | pass | release-candidate workflow smoke test |
| Check | Result | Evidence |
|---|---|---|
| Router contracts | pass | tools/check_router_contracts.py |
| CI matrix policy | pass | tools/ci_matrix.py check |
| Golden vectors | pass | tools/verify_golden_vectors.py |
| Property tests | pass | pytest tests/property --hypothesis-profile=release |
| Artifact | Status |
|---|---|
| SBOM | present |
| SHA256SUMS.txt | present |
Albucore 0.3.1 improves float16/float32 conversion routing and fixes typing and decorator contracts.
Albucore 0.3.1 improves float16/float32 conversion routing and fixes typing and decorator contracts.
Local validation: 3,676 tests passed with the release Hypothesis profile and warnings treated as errors; all pre-commit hooks and 45 router contracts passed.
Release validation: CI, release candidate, and publication. The attached bundle includes the wheel, source distribution, SBOM, checksums, dependency export, candidate metadata, and verification summary. PyPI wheel and source hashes match the candidate; both files have trusted-publishing provenance.
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Area | Result | Evidence |
|---|---|---|
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Headless wheel smoke | pass | release-candidate workflow smoke test |
| Check | Result | Evidence |
|---|---|---|
| Router contracts | pass | tools/check_router_contracts.py |
| CI matrix policy | pass | tools/ci_matrix.py check |
| Golden vectors | pass | tools/verify_golden_vectors.py |
| Property tests | pass | pytest tests/property --hypothesis-profile=release |
| Artifact | Status |
|---|---|
| SBOM | present |
| SHA256SUMS.txt | present |
chore(deps): bump the github-actions group with 2 updates by @dependabot[bot] in https://github.com/albumentations-team/albucore/pull/174
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.20...0.2.21
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Area | Result | Evidence |
|---|---|---|
| Lock freshness | pass | release-candidate workflow uv lock --check |
| Headless wheel smoke | pass | release-candidate workflow smoke test |
| Check | Result | Evidence |
|---|---|---|
| Router contracts | pass | tools/check_router_contracts.py |
| CI matrix policy | pass | tools/ci_matrix.py check |
| Golden vectors | pass | tools/verify_golden_vectors.py |
| Property tests | pass | pytest tests/property --hypothesis-profile=release |
| Artifact | Status |
|---|---|
| SBOM | present |
| SHA256SUMS.txt | present |
feat(geometric): add benchmark-routed 3D constant padding by @ternaus in https://github.com/albumentations-team/albucore/pull/160
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.16...0.2.18
refactor: remove global contiguous output normalization by @ternaus in https://github.com/albumentations-team/albucore/pull/157
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.15...0.2.16
feat(geometric): add dense 3d remapping by @ternaus in https://github.com/albumentations-team/albucore/pull/155
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.14...0.2.15
refactor(ci): adopt shared foundation actions by @ternaus in https://github.com/albumentations-team/albucore/pull/147
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.13...0.2.14
refactor: retire batch-volume target path and document CPU Torch routing by @ternaus in https://github.com/albumentations-team/albucore/pull/140
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.12...0.2.13
feat: add warp_affine3d by @ternaus in https://github.com/albumentations-team/albucore/pull/136
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.10...0.2.12
feat: add mandatory CPU Torch backend and resize3d by @ternaus in https://github.com/albumentations-team/albucore/pull/135
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.9...0.2.10
ci: add NumPy assertion lint hook by @ternaus in https://github.com/albumentations-team/albucore/pull/129
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.8...0.2.9
fix: suppress spurious macOS matmul warnings by @ternaus in https://github.com/albumentations-team/albucore/pull/127
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.7...0.2.8
Fix float64 operand promotion in float32 arithmetic by @ternaus in https://github.com/albumentations-team/albucore/pull/122
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.6...0.2.7
docs(performance): codify optimization workflow by @ternaus in https://github.com/albumentations-team/albucore/pull/114
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.5...0.2.6
Add separate Albucore CLA v1 and legal integrity gates by @ternaus in https://github.com/albumentations-team/albucore/pull/110
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.4...0.2.5
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.3...0.2.4
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.2.3...0.2.4
chore(build): migrate from setuptools to hatchling by @ternaus in https://github.com/albumentations-team/albucore/pull/94
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.1.5...0.1.6
chore: bump version to 0.1.5 by @ternaus in https://github.com/albumentations-team/albucore/pull/93
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.1.4...0.1.5
chore: bump version to 0.1.4 by @ternaus in https://github.com/albumentations-team/albucore/pull/92
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.1.3...0.1.4
Chore/GitHub actions uv node24 by @ternaus in https://github.com/albumentations-team/albucore/pull/87
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.1.1...0.1.2
Full Changelog: https://github.com/albumentations-team/albucore/commits/0.1.0
Full Changelog: https://github.com/albumentations-team/albucore/commits/0.1.0
fix(resize): handle INTER_AREA for 5+ channels, add full interpolatio… by @ternaus in https://github.com/albumentations-team/albucore/pull/83
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.40...0.0.41
fix: strip dst from chunk kwargs in maybe_process_in_chunks by @Dipet in https://github.com/albumentations-team/albucore/pull/80
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.39...0.0.40
Fix/warp affine 3x3 matrix updates by @ternaus in https://github.com/albumentations-team/albucore/pull/79
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.38...0.0.39
Geometric optimizations by @ternaus in https://github.com/albumentations-team/albucore/pull/78
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.37...0.0.38
Nothing published for this version
Added py.typed by @ternaus in https://github.com/albumentations-team/albucore/pull/71
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.35...0.0.36
Make OpenCV an optional dependency by @ternaus in https://github.com/albumentations-team/albucore/pull/70
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.34...0.0.35
Cleaned up python 3.9 by @ternaus in https://github.com/albumentations-team/albucore/pull/68
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.33...0.0.34
Fix in normalize and normalize per image for float32 by @ternaus in https://github.com/albumentations-team/albucore/pull/67
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.32...0.0.33
Speedup normalize by @ternaus in https://github.com/albumentations-team/albucore/pull/64
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.31...0.0.32
Speedup normalize by @ternaus in https://github.com/albumentations-team/albucore/pull/62
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.30...0.0.31
Speedup normalize by @ternaus in https://github.com/albumentations-team/albucore/pull/60
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.29...0.0.30
Simplify grayscale by @ternaus in https://github.com/albumentations-team/albucore/pull/58
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.28...0.0.29
fix function by @trylinka in https://github.com/albumentations-team/albucore/pull/57
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.27...0.0.28
add get_image_shape by @trylinka in https://github.com/albumentations-team/albucore/pull/56
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.26...0.0.27
flags support for get_num_channels by @trylinka in https://github.com/albumentations-team/albucore/pull/55
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.25...0.0.26
Add unit tests for to/from_float for image batches by @martinsbruveris in https://github.com/albumentations-team/albucore/pull/50
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.24...0.0.25
Add support for more than 512 channels in horizontal flip by @ternaus in https://github.com/albumentations-team/albucore/pull/49
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.23...0.0.24
Added batch decorator by @ternaus in https://github.com/albumentations-team/albucore/pull/47
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.22...0.0.23
Add batch transform decorator by @ternaus in https://github.com/albumentations-team/albucore/pull/46
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.21...0.0.22
More inplace by @ternaus in https://github.com/albumentations-team/albucore/pull/45
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.20...0.0.21
Add simsimd by @ternaus in https://github.com/albumentations-team/albucore/pull/42
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.19...0.0.20
Fix CI by @ternaus in https://github.com/albumentations-team/albucore/pull/40
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.18...0.0.19
Added more tests by @ternaus in https://github.com/albumentations-team/albucore/pull/32
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.17...0.0.18
Added decorator to show way to process data by @ternaus in https://github.com/albumentations-team/albucore/pull/31
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.16...0.0.17
Add hflip and vflip by @ternaus in https://github.com/albumentations-team/albucore/pull/30
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.15...0.0.16
Moved to_float and from_float from Albumentations by @ternaus in https://github.com/albumentations-team/albucore/pull/28
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.14...0.0.15
Update package dependencies by @daskol in https://github.com/albumentations-team/albucore/pull/26
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.13...0.0.14
fix test_add by @ayasyrev in https://github.com/albumentations-team/albucore/pull/23
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.12...0.0.13
Added tests to check output shape and dtype by @ternaus in https://github.com/albumentations-team/albucore/pull/20
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.11...0.0.12
Address add negative by @ternaus in https://github.com/albumentations-team/albucore/pull/18
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.10...0.0.11
check len of multiplier and process int mult by @ayasyrev in https://github.com/albumentations-team/albucore/pull/6
Full Changelog: https://github.com/albumentations-team/albucore/compare/0.0.4...0.0.10
Nothing published for this version
Nothing published for this version
One column per month.