rawpipe

A collection of reference ISP algorithms, sufficient for producing a reasonably good looking image from raw sensor data. Each algorithm takes in a frame in RGB or raw format and returns a modified copy of the frame. The frame is expected to be a NumPy float array with either 2 or 3 dimensions, depending on the function. Some of the algorithms can be applied in different orders (demosaicing before or after linearization, for example), but the reference ordering is as shown below.

Example:

import rawpipe
...
raw = rawpipe.linearize(raw, blacklevel=64, whitelevel=1023)
rgb = rawpipe.wb(rgb, [1.5, 2.0], "RGGB")
rgb = rawpipe.demosaic(raw, "RGGB", downsample=True)
rgb = rawpipe.downsample(rgb, iterations=1)
rgb = rawpipe.lsc(rgb, my_vignetting_map)
rgb = rawpipe.lsc(rgb, my_color_shading_map)
rgb = rawpipe.ccm(rgb, my_3x3_color_matrix)
rgb = rawpipe.resize(rgb, 400, 300)
rgb = rawpipe.gamut(rgb, "ACES")
rgb = rawpipe.tonemap(rgb, "Reinhard")
rgb = rawpipe.chroma_denoise(rgb)
rgb = rawpipe.saturate(rgb, lambda x: x ** 0.75)
rgb = rawpipe.gamma(rgb, "sRGB")
rgb = rawpipe.quantize8(rgb)

Installing on Linux:

pip install rawpipe

Installing development dependencies:

uv sync --extra dev

Documentation:

pydoc rawpipe

Building & installing from source:

make install

Running tests locally:

uv run pytest

Building distributions locally:

uv build

Releasing to PyPI:

Publishing is handled by GitHub Actions with PyPI Trusted Publishing.