search

JaneliaSciComp / BigStitcher-Spark

Running compute-intense parts of BigStitcher distributed
BSD 2-Clause "Simplified" License
18 stars 10 forks source link

Speed up SparkAffineFusion #23

Closed tpietzsch closed 4 months ago

tpietzsch commented 4 months ago

Use low-level net.imglib2.algorithm.blocks stuff to improve performance of the default scenario for SparkAffineFusion (3D, with blending weights, without content-based weights).

tpietzsch commented 4 months ago

Forcing Spark to run only one job (with --blockScale 40,40,40), stitching the example dataset (https://github.com/JaneliaSciComp/BigStitcher-Spark?tab=readme-ov-file#examples) takes 8.8s with this PR vs 82.5s before. (both averages over 4 runs on MacBook Air M1)