I have a neural network that takes multiple inputs and produces a semantic segmentation output. I was able to train the network fine, but when I try to test the network performance using the function emanticseg matlab throws this error:
Error using DAGNetwork/createDispatcher (line 364) Input data for a multiple input network must be a combined or transformed datastore.
My input is a combined datastore. I looked up the details of semanticseg and found this:
The read function of the datastore must return a numeric array, cell array, or table. For cell arrays or tables with multiple columns, the function processes only the first column.
Combined data stores have 2 columns, and for a multiple input network it is necessary to read both columns because they correspond to each input layer. So, is it even possible to use semanticseg on a multiple input network? How else can I get my network to segment test images so I can evaluate the performance?
I had to copy the semanticseg function from the matlab library and create my own version of it that can handle combinedDatastores. The predict function that semanticseg uses can use combined datastores, but the wrapper function cannot handle them. So I altered it. If anyone else has the same issue let me know and I can show you what I did.