Trouble Converting IR to Caffe

[lamdawr]

Hi

Platform (like ubuntu 16.04/win10): Ubuntu 14.04

Python version: 3.6 and 2.7

Source framework with version (like Tensorflow 1.4.1 with GPU): Keras 2.1.3

Destination framework with version (like CNTK 2.3 with GPU): Caffe

Pre-trained model path (webpath or webdisk path):

Running scripts:

I get the following error message while trying to create the .caffemodel

File "caffe_model1.py", line 65
print(n.to_proto(), file=fpb)
^
SyntaxError: invalid syntax
root@xxxxxxxxxxxx:/xxx/xxx/xxx/kerastocaffe# python caffe_model1.py -w caffe_model1.npy -p caffe_model1.prototxt -m caffe_model1.caffemodel
File "caffe_model1.py", line 65
print(n.to_proto(), file=fpb)
^
SyntaxError: invalid syntax

Thanks !

[kitstar]

could you add a line in caffe_model1.py

from __future__ import print_function

and try if it works?

[lamdawr]

Thank you very much for the reply.
But I got the following error after adding the line from future import print_function :

Traceback (most recent call last):
File "caffe_model1.py", line 98, in
gen_weight(args.weight_file, args.model, args.prototxt)
File "caffe_model1.py", line 72, in gen_weight
net = caffe.Net(prototxt, caffe.TRAIN)
Boost.Python.ArgumentError: Python argument types in
Net.init(Net, unicode, int)
did not match C++ signature:
init(boost::python::api::object, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, int)
init(boost::python::api::object, std::__cxx11::basic_string<char, std::char_traits, std::allocator > network_file, int phase, int level=0, boost::python::api::object stages=None, boost::python::api::object weights=None)

[kitstar]

hi @lamdawr , Could you share some files for us to reproduce? Thanks.

[lamdawr]

Hi @kitstar , what files would you like ?

[kitstar]

The code and models of source framework would be fine. And you can post the caffe_model1.py for quick check. Thanks.

[galli-leo]

Traceback (most recent call last):
File "caffe_model1.py", line 98, in
gen_weight(args.weight_file, args.model, args.prototxt)
File "caffe_model1.py", line 72, in gen_weight
net = caffe.Net(prototxt, caffe.TRAIN)
Boost.Python.ArgumentError: Python argument types in
Net.init(Net, unicode, int)
did not match C++ signature:
init(boost::python::api::object, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, std::__cxx11::basic_string<char, std::char_traits, std::allocator >, int)
init(boost::python::api::object, std::__cxx11::basic_string<char, std::char_traits, std::allocator > network_file, int phase, int level=0, boost::python::api::object stages=None, boost::python::api::object weights=None)

This issue can be solved by converting the filenames from unicode to string, i.e. caffe.Net(str(prototxt), caffe.TRAIN)

[kitstar]

Fixed in last commit. Thanks!

[galli-leo]

@kitstar Actually, that doesn't work, since the save method also expects a string (and not unicode). I have a fix in my local install, which I can commit with the other caffe emitter changes.