Ref: http://caffe.berkeleyvision.org/gathered/examples/finetune_flickr_style.html
e.g:
Because we are predicting 20 classes instead of a 1,000, we do need to change the last layer in the model. Therefore, we change the name of the last layer from fc8
to fc8_flickr
in our prototxt. Since there is no layer named that in the bvlc_reference_caffenet
, that layer will begin training with random weights.