Caffe hack: outputting the FC7 layer

The Caffe deep learning framework has a nice set of python scripts to help automate classification jobs. However, I found the standard classifier.py output rather limited. The script does not output other data than the predicted result.

Some applications could benefit from outputting the final FC7 classification weights. These weights together with a classification key can then be used to assign different labels (semantic interpretations) using the same classification run.  Below you find my new python (classifier.py) script which outputs the FC7 layer.

This new script allows me to assign classification labels using both the SUN database and the Places205 database in one pass using the MIT places convolutional neural network (CNN).