Trained weights and run outputs for "Curriculum Modality Dropout and Self-Consistency for Robust Optical–SAR Flood Mapping with Missing Sensors"

Trained model weights, run logs and result tables for the manuscript on curriculum modality dropout and self-consistency for optical–SAR flood mapping with missing sensors (Sen1Floods11 and SenForFlood). The package holds exponential-moving-average weights at the selected epoch for 25 run identifiers (3 seeds each: 1, 2, 42), the configuration, per-epoch history, result file and per-chip confusion counts of each run, the result and audit tables, a snapshot of the training and evaluation code, normalisation statistics and the split manifest. A manifest lists the size and SHA-256 of every file, and a coverage report lists the runs included. The datasets themselves are not included. No checkpoint was kept for the FPN row. The code release is archived separately at 10.5281/zenodo.23457175.