/fpzip

Cython bindings for fpzip, a floating point image compression algorithm.

Primary LanguageC++BSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

PyPI version

fpzip

fpzip is a compression algorithm supporting lossless and lossy encoding for up to 4 dimensional floating point data. This package contains Python C++ bindings for the fpzip algorithm (version 1.3.0). The version number for this package is independent. Python 3.7+ is supported.

import fpzip
import numpy as np

data = np.array(..., dtype=np.float32) # up to 4d float or double array
# Compress data losslessly, interpreting the underlying buffer in C (default) or F order.
compressed_bytes = fpzip.compress(data, precision=0, order='C') # returns byte string
# Back to 3d or 4d float or double array, decode as C (default) or F order.
data_again = fpzip.decompress(compressed_bytes, order='C')

Installation

pip Binary Installation

pip install fpzip

If we have a precompiled binary available the above command should just work. However, if you have to compile from sounce, it's unfortunately necessary to install numpy first because of a quirk in the Python installation procedure that won't easily recognize when a numpy installation completes in the same process. There are some hacks, but I haven't gotten them to work.

pip Source Installation

Requires C++ compiler.

pip install numpy
pip install fpzip

Direct Installation

Requires C++ compiler.

$ pip install numpy
$ python setup.py develop

References

Algorithm and C++ code by Peter Lindstrom and Martin Isenburg. Cython interface code by William Silversmith. Check out Dr. Lindstrom's site or the fpzip Github page.

  1. Peter Lindstrom and Martin Isenburg, "Fast and Efficient Compression of Floating-Point Data," IEEE Transactions on Visualization and Computer Graphics, 12(5):1245-1250, September-October 2006, doi:10.1109/TVCG.2006.143.