4

I want to try using pycaffe interface to train caffe models. Particularly I want to use a python layer for input data.

Will this training be slow or will pycaffe just act as an interface to actual caffe?What are the effects?

Any attempts to enlighten the novice are thanked.

Shai
  • 111,146
  • 38
  • 238
  • 371
curio17
  • 660
  • 1
  • 6
  • 15

2 Answers2

2

pycaffe is a Python interface to caffe, so running from pycaffe just calls caffe methods under the hood.

On the other hand, having a Python layer in your model does not compel you to use pycaffe. You can run this model like any other from command line.

Shai
  • 111,146
  • 38
  • 238
  • 371
1

This can be asked as a general question, not only to caffe. What pycaffe does is simply calling the C++ methods that are implemented. Of coruse you will have an overhead to to the python interface and the calling of the function. This blog evaluates the The cost of a Python function call.

Furthermore, if you use a Python IDE like e.g Spyder you give some resources to the IDE. If you run your script in a python-console it may take less resources.

All in all it can be said that if you have a average machine pycaffe will work just as good as the C++ implementation and you will not directly feel any different.

Pycaffe will be easier to handle than the C++ impelementation thanks to abstraction.

Kev1n91
  • 3,553
  • 8
  • 46
  • 96