What version of CellProfiler should I use? Cellprofiler 3.1.9 software# You can download a stable release for macOS and Windows from theĬontributing or planning to contribute to CellProfiler We recommend the stable release of CellProfiler. You can download a beta release for macOS and Windows from theĬreated 22 Nov, 2021 Pull Request #4483 User Lukeozn Let us know if we’ve inadvertently broken your module by You can download a nightly release for macOS and Windows from the Maintainer of a third-party CellProfiler module Instructions for compiling CellProfiler on In using CellProfiler for large images (~8GB) we've found we run out of memory even on our cluster's 1.5TB nodes. The cause of the problem is in MeasureObjectNeighbors. Specifically at the point of the code modified in this PR, the original allocates 4 arrays: i, j, distance_matrix and order. Each of these arrays is of size nobjects*nneighbors. For 8 byte elements, this results in memory usage on the order of 4 * 200000 * 200000 * 8 * 10^-9 = 1280GB. But of these four arrays, the remaining code only makes use of the first three columns of order (the nearest neighbors). We have modified the code to only allocate the memory that is needed, which for the data mentioned above should not need to allocate over 1GB. This code also runs faster than the original as we don't do a full sort on the distances, and instead find just the closest three with argpartition (N^2 rather than N^2*logN complexity). One might worry the explicit Python loop would slow things down, but in testing with random arrays of various sizes we've found about a 10x speed improvement. # An error report file with more information is saved as: To enable core dumping, try "ulimit -c unlimited" before starting Java again # C _ZN33wxMacCoreGraphicsPenBrushDataBase22CalculateShadingValuesEPvPKdPd+0x11f # Java VM: OpenJDK 64-Bit Server VM (25.242-b08 mixed mode bsd-amd64 compressed oops) # A fatal error has been detected by the Java Runtime Environment: The replacement code should be functionally equivalent to the original and in testing on our pipelines we are now able to segment much larger images.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |