Backward compatibility

Small refactoring

• Templated AngularVoxels and HistogramVoxels
• RadField3D Dataset update
  • layer renaming: hits -> flux
  • channel renaming: xray_beam -> 'direct_beam'

fixed a bug in 1.3.1 which broke old files

Adding direction resolved layers

This update introduces a new datatype for voxels.

• SphericalVoxel: Direction resolved binning voxel type
• Updated PyBind11 to version 3.0.3
• Added simple python tool to 3D render a radiation field file (this is only part of the repo and not of the pip release)
• Adding prebuild binaries for python 3.14.x
• Small improvements on file version handling by the library

Bug fixes and introduction of copy functionality

• Fixed bug in get_layer_as_ndarray where a race condition occurred
• Introduced a parameter copy: bool for all get_layer_as_ndarray and get_as_ndarray methods
  • copy == True: A new copy of the fields data buffer is returned (no mutexing)
  • copy == False: A reference on the fields data buffer is returned (This requires mutexing)
• Finally: Use C- instead of Fortran-Ordering when accessing the data buffers from numpy
• Pytorch datasets no longer fully initialize themselves on create, which could lead to race conditions under Windows, but on the fly later on

Full Changelog: 1.2.2...1.2.3

Quick bug fix for rectangle beam params

1.2.2

adding unsigned char and uint8_t as possible datatypes and fix bug wh…

Bug fixes, adding unsigned bytes and radiation source origins

• DType.BYTE now refers to uint8_t instead of char
• Adding DType.SCHAR which refers to a char
• Adding documentation for type handling
• Adding the loading of the radiation source origin

New metadata API

New Metadata API layer for working from python

from RadFiled3D.metadata.v1 import Metadata
from RadFiled3D.utils import FieldStore    # important to use RadFiled3D.utils module, as the plain FieldStore will not load the API

# support of accessing more simulation parameters from RadField3D datasets
m = FieldStore.load_metadata(file)
m.simulation.tube.spectrum
m.simulation.tube.field_shape # [FieldShape.CONE, FieldShape.RECTANGLE, FieldShape.ELLIPSIS]
m.simulation.tube.opening_angle_deg   # for cone fields
m.simulation.tube.field_rect_dimensions_m   # for rect fields
m.simulation.tube.field_ellipsis_opening_angles_deg  # for ellipsis fields

Bug fix for histograms in metadata
Bug fix for numpy accessing of fields switched from fortan-style to C-style

Full Changelog: 1.1.5...1.2.0

Bug fixed

1.1.5

fixing loading of vecs from other compilers

Fixed collating namedtuples

Named tuples with no geometry information could not be batched. This is now fixed by the DataLoader.
Alternative, different NamedTuple classes would have made some things unhandy.

Adding geometry

C++ Backend

• more supported compiler representations of the glm::vec[n] vectors during loading including "struct glm::vec[n]"

Python frontend

• removing metrics tuples as they have no specific use here
• adding the possibility of declaring dataprocessing torch modules for applying dataaugmentation for extending the virtual dataset size
• adding geometry loading from datasets that contain geometry information

Faster Multi-Voxel accessing

It's now possible to request a set of voxels from the same file without reopening it. This is used for the voxelwise pytorch datasets to speed uncached read of single voxels.
Fixed a bug when loading single voxels that resulted in a shift of about 7 voxels per row.