I could be wrong but I think part of the issue is this needs some large files for the trained dataset?