How to view net predictions using only command line tools?

44 views
Skip to first unread message

Bo Moon

unread,
Mar 6, 2016, 11:56:24 AM3/6/16
to Caffe Users
I can use the caffe binary to train and test nets, but I don't see any way to actually view what the net predicts, regardless of whether it's classification or regression. Ideally, I would like to see what the final output blob values are, but even seeing the softmax classification label would be fine. Is there a way to do this using just the caffe tools in the shell? (I ask this because I'm averse to using C++, and building the wrappers raises numerous errors.)

Jan

unread,
Mar 6, 2016, 1:55:31 PM3/6/16
to Caffe Users
Building and using the python wrapper should not be a problem. Usually errors arising there are due to missing dependencies. Since I _never_ used the command line tools for anything other than pure training (because I need more flexibility), I am not very sure how to use them exactly. And reading in an image, feeding it through the network and displaying the outputs with pycaffe is a matter of about 4 lines of code.

Jan

Bo Moon

unread,
Mar 6, 2016, 2:09:10 PM3/6/16
to Caffe Users
Yeah, after looking through the Python examples on the site, it's an extremely attractive option I want to use, but I get an error of "TypeError: No to_python (by-value) converter found for C++ type: boost::shared_ptr<caffe::Blob<float>>", which apparently was fixed recently. However, if I use the newest build, I'm no longer able to even build the caffe binary, let alone pycaffe, so this isn't an option until I get that issue resolved. But that's all a separate matter--long story short, I can't build the latest Caffe until I get the issue resolved, so I would like to have a non-programmatic way of seeing prediction results as a workaround, otherwise I'll have to wait until I can get help on a different issue.
Reply all
Reply to author
Forward
0 new messages