import compressed_segmentation as cseg
sx, sy, sz = (128,128,128)
dtype = np.uint64
order = 'C'
labels = np.arange(0, sx*sy*sz, dtype=dtype).reshape((sx,sy,sz), order=order)
compressed = cseg.compress(labels, order=order)
recovered = cseg.decompress(
compressed, (sx,sy,sz) dtype=dtype, order=order
)
arr = CompressedSegmentationArray(
compressed, shape=(sx,sy,sz), dtype=dtype
)
label = arr[54,32,103] # random access to single voxels w/o decompressing
uniq_labels = arr.labels() # get all distinct values w/o decompressing
binary2 = arr.remap({ 1: 2 }, preserve_missing_labels=False) # remap labels in segmentation w/o decompressing
recovered = arr.numpy() # decompress to a numpy array, same as decompress
124213 in arr # test if a value is in the array
cseg compress connectomics.npy cseg decompress connectomics.npy.cseg --volume-size 512,512,512 --bytes 4
NOTE: This repository is the PyPI distribution repo but is based on work done by Jeremy Maitin-Shepard (Google), Stephen Plaza (Janelia Research Campus), and William Silversmith (Princeton) here: https://github.com/janelia-flyem/compressedseg
This library contains routined to decompress and compress segmentation and to manipulate compressed segmentation data defined by the neuroglancer project. compressed_segmentation essentially renumbers large bit width labels to smaller ones in chunks. This provides for large reductions in memory usage and higher compression.
Note that limitations in the compressed_segmentation format restrict the size of the chunk that can be compressed. As this limitation is data dependent, for example a random array with 1024 labels passes testing at 256x256x128, but 256x256x256 often does not.
Compiling as a shared library. Feel free to subsititute e.g. clang for the C++ compiler.
g++ -std=c++11 -O3 -fPIC -shared -I./include src/compress_segmentation.cc src/decompress_segmentation.cc -o compress_segmentation.so
$ pip install compressed-segmentation $ python >>> import compressed_segmentation as cseg >>> help(cseg)
If there are pre-built binaries available for your architecture this should just work.
If you need to build from source, you will need to have a C++ compiler installed:
$ sudo apt-get install g++ python3-dev $ pip install numpy $ pip install compressed-segmentation $ python >>> import compressed_segmentation as cseg >>> help(cseg)
Requires a C++ compiler such as g++ or clang.
Works with both Python 2 and 3. Encodes from / decodes to 3D or 4D numpy ndarrays.
$ sudo apt-get install g++ python3-dev $ pip install -r requirements.txt $ python setup.py install $ python >>> import compressed_segmentation as cseg >>> help(cseg)
Please see the licenses in this repo.
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4