Skip to content

Install fuse-augmentations

The base package requires Python 3.10 or newer and PyTorch 2.2 or newer. Kornia, TorchVision, and Albumentations are optional because the native builder can create a useful pipeline without them.

!!! note "Project maturity" The package is currently classified Alpha. Pin versions in production or research environments and validate the exact pipeline after upgrades.

Base installation

python -m pip install fuse-augmentations

This installs NumPy, PyTorch, and the package itself. It is enough for Compose.from_params.

Optional backends

Install only the adapter ecosystems you use:

python -m pip install "fuse-augmentations[kornia]"

python -m pip install "fuse-augmentations[torchvision]"

python -m pip install "fuse-augmentations[albumentations]"

python -m pip install "fuse-augmentations[all]"

The extras enable adapter support; they do not make every upstream transform or parameter combination fusible. Check the capability tables.

Verify the installation

python -c "import torch, fuse_augmentations"

The same smoke check is executable in the generated documentation test suite:

import torch
import fuse_augmentations

assert torch.__version__
assert fuse_augmentations.__version__

Both import namespaces expose the same objects:

from fuse_aug import Compose as ShortCompose
from fuse_augmentations import Compose

assert ShortCompose is Compose

Long-form documentation uses fuse_augmentations so the import matches the distribution name.

Build these docs locally

The repository uses uv for its locked development environment:

uv sync --group docs
uv run --group docs mkdocs serve

Run the release-style documentation gate with:

uv run --group docs mkdocs build --strict

The generated site is written to site/. The docs dependency group is not installed for package users.