Merging different input channels into coffee?

Question

I would like to process different data types first, and then merge them into a common layer. Is this possible in Caffe, and if so, what would be the best way to do this?

I read that you can define multiple data layers in one prototype file. But how to fuse them?

Can I create an InnerProduct layer and specify multiple bottom layers? Or do I need to merge the individual layers first using the Concat layer?

For any small code example, I would be very grateful!

Answer 1

As discussed in the comments above, InnerProduct works with a single input. Then merging (concatenation) can be performed at a specific Concat level with this configuration:

 layer { name: "concat" bottom: "in1" bottom: "in2" top: "out" type: "Concat" concat_param { axis: 1 } } 

The official documentation contains more detailed information about this layer: http://caffe.berkeleyvision.org/tutorial/layers.html