Running compute-intense parts of BigStitcher distributed. For now we support fusion with affine transformation models (including translations of course). It should scale very well to large datasets as it tests for each block that is written which images are overlapping. You simply need to specify the XML of a BigSticher project and decide which channels, timepoints, etc. to fuse. Warning: not tested on 2D yet.
Sharing this early as it might be useful ...
Here is my example config for this example dataset for the main class net.preibisch.bigstitcher.spark.AffineFusion:
-x '~/test/dataset.xml'
-o '~/test/test-spark.n5'
-d '/ch488/s0'
--UINT8
--minIntensity 1
--maxIntensity 254
--channelId 0
Note: here I save it as UINT8 [0..255] and scale all intensities between 1 and 254 to that range (so it is more obvious what happens). If you omit UINT8, it'll save as FLOAT32 and no minIntensity and maxIntensity are required. UINT16 [0..65535] is also supported.
Importantly: since we have more than one channel, I specified to use channel 0, otherwise the channels are fused together, which is most likely not desired. Same applies if multiple timepoints are present.
The blocksize is currently hardcoded to 128x128x128, but can easily be added as another parameter (pull requests welcome :).
And for local spark you need JVM paramters (8 cores, 50GB RAM):
-Dspark.master=local[8] -Xmx50G
Ask your sysadmin for help how to run it on your cluster. mvn clean package -P fatjar builds target/BigStitcher-Spark-0.0.1-SNAPSHOT.jar for distribution. Importantly, if you use HDF5 as input data in a distributed scenario, you need to set a common path for extracting the HDF5 binaries (see solved issue here), e.g.
--conf spark.executor.extraJavaOptions=-Dnative.libpath.jhdf5=/groups/spruston/home/moharb/libjhdf5.so
You can open the N5 in Fiji (File > Import > N5) or by using n5-view from the n5-utils package (https://github.com/saalfeldlab/n5-utils).
You can create a multiresolution pyramid of this data using https://github.com/saalfeldlab/n5-spark
Update: now there is support for non-rigid distributed fusion using net.preibisch.bigstitcher.spark.NonRigidFusionSpark
In order to run it one needs to additionally define the corresponding interest points, e.g. -ip beads that will be used to compute the non-rigid transformation.