I checked the fine tuning tutorial on Caffe's site, but I'm still confused. Lets say you have a data set of dogs and cats, and want to use one of the prebuilt networks like AlexNet with Caffe. How does this work? Do you just switch out the last layer with something else? Right now the prebuilt ones will make predictions with the 1000 categories from Imagenet, how do you make it predict on your set, like 'cat' or 'dog'?
Has anyone seen a practical tutorial for something like this, with code? Thanks!
[link][4 comments]