scitomo/walnut-edip-sparse20

Scitomo Walnut EDIP Sparse-20 prepared dataset

This is a derived Scitomo Sparse-20 preparation of the public Walnut-1 cone-beam X-ray CT acquisition. It is not the original Walnut archive, not a published EDIP reconstruction, and not a blessed Scitomo result. The package contains measured projections, corrected vector-cone geometry, and the published AGD-50 evaluation reference used by the maintained Scitomo Walnut EDIP evidence workflow.

Provenance and attribution

The source authority is the Scitomo catalog entry walnut_1_zenodo, pinned to Zenodo record 2686726, DOI `10.5281/zenodo.2686726`, and the single provider member Walnut1.zip.

The original dataset is described in the Scientific Data publication `10.1038/s41597-019-0235-y`. The official reconstruction-code reference is `cicwi/WalnutReconstructionCodes`. The source record declares CC BY 4.0 licensing. Please retain this attribution when using or redistributing the derived preparation.

The exact provider identity used for preparation is:

provider file: Walnut1.zip
byte size: 6058580169
MD5: 0ecd525f245e3029788538a44f5491e3
SHA-256: ae47f2b2cf19b34191ee0605e3436451637779e5d41b2d178df385f524064392

The preparation was rerun from that pinned source with Scitomo commit e568717442680afd45437dd713feda04e73a244f and tools/datasets/prepare_walnut_edip.py blob ac1a71bfa14358a492fc637f95e84dd6a4334584. The fresh replay was compared byte-for-byte with the reviewed four-file package before publication staging.

Exact Scitomo preparation semantics

The preparation is the existing G9/0091-0094 Scitomo implementation. No alternate preprocessing, geometry reconstruction, angle convention, or fallback backend is used.

• —Walnut-1, orbit 2, with 1200 unique views.
• —The 20 measured views use corrected geometry rows 0,60,...,1140 and scanner TIFF members 1200,1140,...,60.
• —Each detector image uses the existing CWI orientation transpose(flipud(raw)).
• —The dark field is di000000.tif; the flat field is the float32 mean of io000000.tif and io000001.tif.
• —Calibration is exactly -log((scan-dark)/(flat-dark)), with no clipping or epsilon substitution. Invalid non-positive calibration values fail closed.
• —Detector reduction is centered stride-3 decimation, using zero-based rows 1,4,...,970 and columns 1,4,...,766. Projections are canonical Scitomo (A,V,U)=(20,324,256) float32 arrays.
• —cone_vec contains the exact 20 corrected vector-cone rows as float64 millimetre geometry. Its detector basis vectors are scaled by 3 for the centered stride-3 detector grid. cone_vec is the pose authority.
• —Nominal labels are 0,18,...,342 degrees from the documented 0.3-degree increment. They are ordered labels only and are not used to rebuild or reinterpret vector poses.
• —The reference is full_AGD_50_{slice:06d}.tiff for slices 000000 through
• —It is independently reduced from (501,501,501) to (167,167,167) by exact centered 3x3x3 mean pooling, producing float32 data on the reviewed centered 0.3 mm isotropic grid.

This package stops at prepared measured inputs and evaluation reference. It contains no DIP/EDIP reconstruction, pretrained parameters, CUDA execution, acceptance threshold, visual review, or blessing metadata.

Files and identities

The filenames below are the manifest-owned names; they are intentionally not renamed to generic aliases.

The manifest is the package contract. It records the source authority, procedure, shapes, dtypes, nominal labels, pose-authority declaration, and artifact SHA-256 values.

Loading in Scitomo

Use the repository-owned preparation loader against a directory containing the four files. It verifies the manifest authority and every array's path, hash, shape, and dtype before constructing the measured Problem:

from pathlib import Path
from tools.datasets.prepare_walnut_edip import load_walnut_edip_sparse20_problem

problem = load_walnut_edip_sparse20_problem(
    prepared_dir=Path("/path/to/this/dataset"),
)