To be able to use a Caffe model in Caffe2, we need to convert it from its Caffe formats to Caffe2 file formats. Caffe2 provides a script named python/caffe_translator.py that can be used for this purpose.
For example, we can convert our AlexNet files from Caffe to Caffe2 by invoking the script as follows:
$ python python/caffe_translator.py path_to_caffe/models/bvlc_alexnet/deploy.prototxt path_to_caffe/models/bvlc_alexnet/bvlc_alexnet.caffemodel --init_net init_net.pb --predict_net predict_net.pb
Running this script generates three files, predict_net.pb, predict_net.pbtxt, and init_net.pb, for AlexNet:
Figure 4.3: AlexNet network structure in Caffe2
Figure 4.3 shows the AlexNet network structure in Caffe2 after it was converted from the Caffe model. This graph visualization was generated using the Caffe2 net_drawer.py tool that utilizes GraphViz...
