after `--allfolds` training.py, there is no results but a caffemodel: ****