First slices from beamline exports

These recipes use the public Python API with an installed TomoJAX environment. Run Python and tomojax from that environment; a source checkout is not required. If your environment is .venv, activate it with source .venv/bin/activate. Start with a single detector row so the preview does not allocate a full volume. The examples are raw counts, not already-log-transformed projections.

The calibration numbers below belong to the named acquisitions. For another scan, use its own recorded calibration; do not copy a plausible pixel size or rotation centre. File recognition and physical calibration are separate: an otherwise recognised file can still need detector= or angles=.

ESRF NXtomo exportCopy Anchor Link

For the bamboo scan exported from Bliss by nxtomomill, use nabu's public test file. This server currently has an invalid TLS certificate; the supplied link is HTTP. Do not disable certificate verification for other downloads.

import tomojax as tj

frames = tj.load_frames("bamboo_reduced_bliss.nx")

The reader finds the group declaring NXtomo, regardless of the entry's name, and reads its sample angles, image keys, flat/dark fields and unit-labelled pixel pitches. If the axial centre is absent, loading estimates it from corrected near-opposite views and records that it was inferred; an explicit calibrated detector= overrides this estimate.

Continue with the one-row preview below. An ESRF NXtomo export is not a raw Bliss master or an EDF folder. Those have different metadata requirements; complete real-scan validation currently covers the NXtomo exports, not every native ESRF layout.

PSI TOMCAT Data Exchange exportCopy Anchor Link

The TomoBank catalogue identifies the tomo_00069 skin dataset (file tomo_00069.h5). It contains counts, flats, darks and /exchange/theta, but no pixel pitch. Its accompanying calibration gives 3.7 µm pixels and rotation-axis column 515.5 on the 1024-column detector. Supply that calibration explicitly, here in millimetres:

import tomojax as tj

pitch = 3.7e-3  # millimetres; this value is not in the HDF5 file
axis_column = 515.5
frames = tj.load_frames(
    "tomo_00069.h5",
    detector=tj.Detector(
        1024, 428, pitch, pitch,
        center=((1023 / 2 - axis_column) * pitch, 0),
    ),
)

Detector.center is a physical offset, not an axis column index. For an axial detector, its horizontal offset is (nu - 1) / 2 - axis_column, multiplied by the pitch. The explicit detector is retained rather than automatically re-estimated. A metadata-only inspection of the original file cannot supply its missing pitch; inspect the calibrated preview saved below instead.

This reads the Data Exchange conversion, not TOMCAT's native GigaFRoST format.

Elettra SYRMEP Data Exchange exportCopy Anchor Link

The TomoBank catalogue identifies the tomo_00024 polystyrene dataset (file tomo_00024.h5). It records its 2.04 µm pixel pitch. The accompanying calibration gives rotation-axis column 972 on 2048 columns:

import tomojax as tj

pitch = 2.04e-3  # millimetres
frames = tj.load_frames(
    "tomo_00024.h5",
    detector=tj.Detector(
        2048, 2048, pitch, pitch,
        center=((2047 / 2 - 972) * pitch, 0),
    ),
)

The reader uses the recorded /exchange/theta array, not the inconsistent angular_step summary in this export. This is local tomography: the sample extends beyond the field of view. A successful preview or centre estimate does not remove truncation artefacts or establish quantitative attenuation. This recipe does not establish support for every native SYRMEP layout.

Correct, reconstruct and inspectCopy Anchor Link

Save one of the three frames = ... snippets above in preview.py and append the following block. It retains one native detector row, applies flat/dark correction and the logarithm, then uses the default axial reconstruction:

row = frames.counts.shape[1] // 2
scan = frames.cropped(slice(row, row + 1), slice(None)).corrected()
result = tj.reconstruct(scan)
tj.save("preview.nxs", result)

Use a new output path, then inspect it from the same environment:

python preview.py
tomojax inspect preview.nxs --preview preview-images

The saved file retains the calibrated geometry, corrected projections and volume. The PNGs are display-scaled previews; use the floating-point data for quantitative work. The Python round trip is:

restored = tj.load_reconstruction("preview.nxs")
print(restored.volume.shape, restored.scan.grid, restored.scan.detector)

For rings, phase retrieval, larger row windows and explicit reconstruction methods, continue with the quickstart. Do not apply another logarithm to the saved corrected projections.

Diamond I13 NXtomo acquisitionCopy Anchor Link

The unpublished Diamond I13 scan 159815 is a GDA NXtomo file, diamond-i13-mg31912-1/raw/159815.nxs, linked to pco1-159815.hdf beside it. Keep both files together. Loading discovers its 3001 sample views, 200 flats in before/after sets, 100 darks and recorded angles without a TIFF export. The detector pixel pitch is not recorded, so tj.load reports the missing calibration: supply detector= through tj.load_frames.

This preview uses pixel units, not millimetres or micrometres. The Savu processing record supplies the crop 441:1872 and axis column 718.5 within that crop: original axis column 1159.5, hence detector offset 120. These are values for this scan, not I13-wide defaults; replace them with your scan's calibration. The acquisition's recorded angle signs are preserved. Negate them explicitly only when comparing to a pipeline using the opposite rotation convention.

import tomojax as tj

frames = tj.load_frames(
    "diamond-i13-mg31912-1/raw/159815.nxs",
    detector=tj.Detector(2560, 2160, 1, 1, center=(120, 0)),
)
scan = frames.cropped(slice(1080, 1081), slice(441, 1872)).corrected()
result = tj.reconstruct(scan)
tj.save("i13-preview.nxs", result)

Save this block as i13_preview.py and run it:

python i13_preview.py
tomojax inspect i13-preview.nxs --preview i13-preview-images

This is the default reconstruction, not a reproduction of Savu's ring-removal/filtering pipeline. For TIFF-only acquisitions without NXtomo metadata, see preparing TIFF data.