Data and code for: Cross-dataset evaluation in pistachio leaf disease diagnosis and the dependence of reported accuracy on the annotation protocol

This deposit contains the derived data, imagery and analysis code required to reproduce every number reported in the accompanying manuscript. Included: the reduced common feature space on which the cross-dataset transfer experiment was run for both assessor-scored datasets; out-of-fold predictions for every classifier under every resampling scheme; repeated cross-validation and equivalence testing; symptom prevalence by orchard; bootstrap confidence intervals for the headline metrics; the photometric robustness analysis; 67 uncontrolled orchard photographs with EXIF stripped; 318 segmented leaf instances; pixel-level lesion masks for 103 annotated leaves; and all analysis scripts. Not included: the raw assessor-scored tabular records of dataset II-b. These were supplied through the regional chamber of agriculture, which does not issue permission for its records to be transferred to third parties. The derived common-space matrices included here are binarised and rank-transformed, so the raw values cannot be reconstructed from them. See README.md inside the archive for a file-by-file description.