Reloading a paused training session from snapshots in PyCaffe. The .solverstate goes where?

61 views

Skip to first unread message

9th Dimension

unread,

Jun 10, 2016, 3:35:02 PM6/10/16

to Caffe Users

I have a very simple script for fine-tuning BVLC.

import caffe

#net = convnet = caffe.Net('train_val.prototxt', 'bvlc_reference_caffenet.caffemodel', caffe.TRAIN)
net = convnet = caffe.Net('train_val.prototxt', 'caffenet_train_snapshot_iter_175.caffemodel', caffe.TRAIN)

solver = caffe.get_solver('adam_solver.prototxt')

solver.solve()

I had to pause training and I'd like to resume. I know that my caffe.Net() object has to be constructed with the snapshot.caffemodel, but where do I feed in the .solverstate file, containing the gradients?

Reply all

Reply to author

Forward

0 new messages