Normalizing the images to OME-* would be helpful in simplifying some aspects of the pipeline code and also eliminating the need for those early modules to include bioformats, but the cost of creating a full copy of the raw input image data is pretty brutal for large cohorts (10s of terabytes of raw data). For example our in-house HPC cluster offers 25 TB scratch space (up from 10 TB last year) which is almost too small for some of our experiments already.