I try to use pre-trained model (VGG 16) to DIGITS for my dataset which has two classes. but I got wrongly training. (I change the name of last layers to fc8_new)
I uploaded deploy.prototxt
and VGG_ILSVRC_16_layers.caffemodel
and synset_words.txt
successfully into DIGITS and tested with my data-set which has two classes.
A deploy.prototxt
won't include any loss or accuracy layers. I'm not really sure what Caffe will do when you try to train a network without any loss layers.
As I answered here, you're going to need to create an "all-in-one" network which includes all the layers needed for training, validation and deployment in a single network. @Shai linked to this helpful answer as well.
As explained on our GitHub README, we prefer to answer these questions on our user group instead of here on SO.