We are using Movidius NCS stick for deep neural network computing. We have successfully run the given examples.
We trained a convolutional neural network(CNN) on the Nvidia GPU using the DIGITS platform and got the caffe
model. Can this caffe model be directly used on/with the NCS without modifications? Using the tool mvNCCompile.pyc, we
have been able to generate the graph file. What file needs to be used as stat.txt for our CNN?
Thanks in advance.
Hello, as I see for all three example networks AlexNet, GoogLeNet and SqueezeNet is used exactly the same stat.txt so why don't you try just to copy it from there. The file contains these 6 floats
0.40787054 0.45752458 0.48109378
0.00392157 0.00392157 0.00392157
Archana, were you able to get the DIGITS output to work with the NCS? I'm having difficulties at the moment. I have a previous model I trained via DIGITS and when I do the following
mvNCCompile deploy.prototext -w snapshot_iter_11970.caffemodel -s 12
I get the following error
Error importing caffe
Where you able to make this work? Any insight would be much appreciated.