fim (FastImage)

A high-performance C++20 library for tile-based image processing using CRTP (Curiously Recurring Template Pattern), with Python bindings that provide a PyVips-style lazy-evaluation pipeline.

This repository is automatically synced from the AI for Oncology monorepo. It builds both with Bazel (via bzlmod) and with Meson (used to produce the Python wheels).

Features

• Tile-based processing: efficient processing of large images through tiling
• CRTP pipeline: zero-overhead composition using compile-time polymorphism
• Lazy evaluation: operations are deferred until a sink materializes the result
• Multiple sources: TIFF (libtiff), PNG (lodepng/libspng), and FastSlide (whole slide images)
• Operators: crop, paste, downsample, resize (Lanczos / Magic2021 kernels), and stack
• Flexible sinks: tiled TIFF/PNG output or zero-copy NumPy arrays
• Python bindings: full Python API via nanobind

Installation (Python)

pip install fimage-python

import fim

img = fim.Image.from_libtiff("input.tiff")
img.crop(100, 100, 512, 512).resize(256, 256).write_png("output.png")

Building from source

Bazel

# Build the C++ library
bazelisk build //:fimage

# Build the Python extension
bazelisk build //python:fim

# Run the test suite
bazelisk test //...

Meson (wheels)

# Build a wheel + sdist for the current interpreter
uv build

The Meson build fetches every dependency through subprojects/*.wrap and links them statically, producing a self-contained _fim extension.

License

Apache 2.0 - see LICENSE for details.