Runs in a browser, or install it offline
The ipywidgets / JupyterLab front end is hosted on the TORCH computing platform, so a browser is enough; Windows / Linux desktop packages are also provided.
X-ray CT data processing and reconstruction: projection pre-processing, phase retrieval, artifact correction, image stitching, and GPU-accelerated / multi-GPU tomographic reconstruction.

HEPSCT is data processing and three-dimensional reconstruction software for X-ray CT. Its front end is built on the JupyterLab ecosystem with ipywidgets, and its back end integrates MOCUPY, the GPU reconstruction core developed by the HEPS X-ray microscopy imaging beamline team (CUDA + Python). It runs in a browser without local installation, and is also shipped as Windows / Linux desktop packages.
Input: uncompressed TIFF (8-bit uint, 16-bit uint, 32-bit float) or HDF5 in the HEPS imaging-beamline layout; output: 32-bit float TIFF slices
TIFF index numbers must have a fixed width, e.g. tomo_0001 … tomo_0180
The reconstruction core is fully GPU accelerated, with multithreaded I/O and overlapped CPU/GPU processing
Automatic multi-scale rotation-axis correction and wavelet-based ring artifact removal
Grid reconstructs a 2k×2k projection dataset in seconds
Multi-GPU parallel reconstruction for large batches of CT data
Projection stitching before reconstruction: adjacent scans are registered on their overlap and then merged at the measured positions; the horizontal, 360°→180° and vertical steps can be run one by one or in a single batch
The result is saved as HDF5 with positions files (recording each tile) and reports; the overlap percentage in the UI only narrows the search range










The ipywidgets / JupyterLab front end is hosted on the TORCH computing platform, so a browser is enough; Windows / Linux desktop packages are also provided.
The reconstruction core is fully GPU accelerated with multithreaded I/O and overlapped CPU/GPU work; Grid reconstructs a 2k×2k projection dataset in seconds.
Automatic multi-scale rotation-axis correction, interpolated between separately measured upper and lower axes, plus wavelet-based ring artifact removal.
Manual and automatic (marker-based) modes correct the projection drift caused by stage jitter.
Horizontal, 360°→180° and vertical stitching merge several detector scans into one volume, with batch runs, positions files and reports.
Reconstruction jobs can be spread over several GPUs, which suits large batches of CT data.
| Version | Platform | File size | Published | Downloads | |
|---|---|---|---|---|---|
| No package yet — use it online in TORCH. | |||||
Contact the maintainers above for usage, adaptation or customisation.