where is the trained model stored after training. Can I use the snapshots such as flownet_iter_xxxx.caffemodel for testing the net?