HTTomo documentation

HTTomo is a Python framework for high-performance tomographic data processing. It coordinates distributed I/O and CPU/GPU processing through MPI and runs pipelines described in readable YAML files.

Where should I start?

New to HTTomo?

Follow the 10-minute quickstart to download a small test dataset, validate a pipeline and run it from start to finish.

Preparing your own data?

Read Loading data, including how to create an HTTomo-compatible NXtomo file.

Building a pipeline?

Browse the ready-to-use pipelines, or configure one from available method templates.

Developing HTTomo or a method?

Start with Architecture and internals, then follow the contribution path that matches your change.

Introduction

• About HTTomo
• How HTTomo runs a pipeline
  • Creating sections
  • Executing a section
  • Moving between sections
  • Completing the pipeline
• Core concepts
  • Pipelines and templates
  • Chunks and blocks
  • Sections
  • Re-slicing

Getting started

• Installation
  • Choose an installation path
  • Conda environment with GPU support
  • CPU-only Conda environment
  • Virtual environment
  • Verify the installation
  • Installation on Other Platforms
• 10-minute quickstart
  • Prerequisites
  • Download the data and pipeline
  • Validate the pipeline
  • Run the reconstruction
  • Next steps
• Run your first pipeline
  • Required inputs
  • Choose an example pipeline
  • Validate the pipeline
  • Run HTTomo
  • Run in parallel
  • Find and inspect the output
• Running HTTomo at Diamond
  • Running HTTomo in parallel
  • Running HTTomo serially on workstation

User guide

• Loading data
  • Standard tomography loader
  • Darks and flats
  • Rotation angles
  • Previewing
  • Continuous-scan subsets
  • Creating an NXtomo file
• Pipeline features
  • Configure a pipeline
  • Optimise pipeline performance
  • Memory and performance
  • Side outputs
  • Centre of Rotation
  • Parameter Sweeping
  • Save intermediate datasets
  • Padding
• Tutorials
  • Real data processing
  • Generate synthetic data
• Troubleshooting
  • Pipeline validation fails
  • HTTomo cannot find NXtomo data
  • MPI or parallel HDF5 fails at startup
  • CUDA or CuPy cannot see a GPU
  • GPU memory is exhausted
  • Pipeline and method templates do not match
  • The output directory cannot be created or written
  • Re-slicing is unexpectedly slow
  • Finding more information
• Frequently asked questions
  • YAML and pipelines
  • Using and extending HTTomo

Pipelines and methods

• Ready-to-use pipelines
  • GPU pipelines
  • Tutorial pipelines
  • Diamond-specific pipelines
  • Parameter-sweep pipelines
  • CPU pipeline
• Versioned downloads
• Choose a pipeline
• Available methods
  • Current method reference
• Processing libraries
  • HTTomolibGPU
  • HTTomolib
  • TomoPy
  • Adding another method or library
• Reconstruction ecosystem
  • Pathways at a glance
  • The GPU pathway
  • The CPU pathway
  • Choosing and troubleshooting a pathway

Reference

• Command-line interface
  • The check command
  • The memory-check command
  • The run command
• Pipeline file reference
  • YAML formatting
  • Method entries
  • Side-output references
  • Parameter sweeps
  • Minimal pipeline
• Run output, logs and monitoring
  • Intermediate HDF5 files
  • Image output and sweeps
  • Snapshots
  • Monitoring output
  • Common log messages
• Version compatibility
  • Backend packages
  • Release information
• Glossary

Developers

• Architecture and internals
  • Execution path
  • Repository responsibilities
  • Public entry points
• Development setup and testing
  • Create a development checkout
  • Run the test suite
  • GPU tests
  • Small-data pipeline tests
  • Run code-quality checks
  • Build the documentation
• How to contribute
  • 1. Put the code in the right project
  • 2. Confirm that HTTomo can call it
  • 3. Register it in httomo-backends
  • 4. Test the integration
  • 5. Submit linked changes
• Adding a processing method
• Integrating methods with httomo-backends
  • Where httomo-backends fits
  • Before starting
  • Integration workflow
  • Common mistakes
• Memory estimation and block sizing
  • How HTTomo selects a block length
  • Where the configuration lives
  • Choose an estimation strategy
  • Direct estimates
  • Module estimates
  • Iterative estimates
  • Data types and allocations
  • Testing an estimator
• Profiling and tracing
  • Record a serial run
  • Record an MPI run
  • Combine and inspect traces
• API
  • Module overview
  • Generated reference