Implement CNN Triplet training.

[paroj]

same as #3123 but rebased on master and including the requested changes.

[soulslicer]

Good day,

I am trying to use your triplet loss layer, but I can't understand the input and output. I currently have a input of [N*3 * 1024] where the data is stored in

[+ 1]
[+ 1]
[- 1],
[+ 2]
[+ 2]
[- 2]
etc.

How can I use the triplet loss layer to minimize this?

Also, why are there two inputs to triplet loss. Shouldnt the feature vectors be enough. the 2nd input is not even used except in:

// CHECK_EQ(bottom[1]->channels(), 1);
// CHECK_EQ(bottom[1]->height(), 1);
// CHECK_EQ(bottom[1]->width(), 1);