Virtual detectors

Virtual detectors#

BF, DF, ADF, arbitrary detector masks and diffraction reductions operate on the same acquisition but can touch very different numbers of detector pixels. Record the mask, original scan/detector shape, representation and reduction dtype when comparing implementations.

Evidence and implementation#

  • Seven ANS acquisitions on CUDA: native-resolution BF/DF/ADF updates, exact parity and corruption recovery.

  • Virtual-image validation record: retained CUDA, Metal and WebGPU gates, timings and unresolved memory cases. Its implementation-status statements belong to the recorded work, not an automatic current-release guarantee.

  • Scientific operation and source map: detector coordinates, masks and optimized reduction ownership.

  • Detector API: public inputs and outputs.

  • Cross-backend parity: exact integer-product versus floating-point requirements.

CoM and DPC have their own entry point because moment fields, rotation and phase integration need additional numerical checks. Full source loading is tracked under IO, not included implicitly in a resident detector-update time.