Code, splits, predictions and analysis outputs for "When enhancement deletes information: a controlled protocol for evaluating intensity preprocessing before CNN classification of medical images"
Annotatsiya
Supporting material for the article "When enhancement deletes information: a controlled protocol for evaluating intensity preprocessing before CNN classification of medical images". This deposit contains everything needed to recompute every number, table and figure in the article and its supplementary material: the full experiment and analysis code, the de-duplicated data splits as image identifiers, the per-image test-set predictions of all 336 training runs, the per-run result files, and the analysis outputs. No dataset images are included. The four datasets analysed (APTOS 2019, Eye Diseases Classification, ISIC 2019 and a parasite microscopy set) remain with their providers under their own terms; this deposit holds only identifiers, model outputs and derived statistics. The included verification script reproduces 324 published numbers directly from this deposit and reports no discrepancies.
Hali tarjima qilinmagan