Skip to first unread message
kys...@gmail.com
Feb 26, 2018, 10:29:08 AM2/26/18
to Caffe Users
Hi,
When I use models/bvlc_googlenet/train_val.prototxt and solver.prototxt to train googlenet from scratch on imagenet, I only got accuracy of 59% at top-1 and 83% at top-5 at max_iteration (10000000) . However, googlenet is claimed to achieve 68.9% and 89.04%. What accuracy did you get? Should I do some modifications to the prototxt files?
Thanks!
When I use models/bvlc_googlenet/train_val.prototxt and solver.prototxt to train googlenet from scratch on imagenet, I only got accuracy of 59% at top-1 and 83% at top-5 at max_iteration (10000000) . However, googlenet is claimed to achieve 68.9% and 89.04%. What accuracy did you get? Should I do some modifications to the prototxt files?
Thanks!
Reply all
Reply to author
Forward
0 new messages