About HTTomo

HTTomo is a Python framework for high-throughput processing and reconstruction of parallel-beam tomography data. It is developed at Diamond Light Source to support increasing data rates, including those anticipated from the Diamond-II upgrade.

HTTomo can divide large three-dimensional datasets into memory-sized chunks and process them in parallel across CPUs or multiple GPUs. Performance comes from distributed I/O, MPI-based in-memory operations, GPU-accelerated methods and CuPy device-to-device processing.

The framework orchestrates methods supplied by external CPU and GPU processing libraries rather than implementing the scientific methods itself. Users combine those methods in reusable YAML pipelines.

Fig. 1 Parallel-beam projection data can be divided between several workers for processing and reconstruction.