I've been testing Calculate Cloud Mask Using Fmask tool in ENVI 5.4 on Sentinel-2 and Landsat-8 data and results have been unsutisfactory. Algorithm tends to overestimate cloud cover and classify bright soils, snow, artificial structures and sometimes even vegetation as clouds. Futhermore, shadows stay undetected, while original algorithm detects cloud shadows as well. Additionally, different configuration parameters doesn't have any affect on results. Have somebody expirienced the same problems?