MatConvNet: The FILTERS depth does not divide the DATA depth

Question

I have the following issue:

---

Error using vl_nnconv
The FILTERS depth does not divide the DATA depth.

Error in dagnn.Conv/forward (line 16)

outputs{1} = vl_nnconv(...

Error in dagnn.Layer/forwardAdvanced (line 85)

outputs = obj.forward(inputs, {net.params(par).value}) ;

Error in dagnn.DagNN/eval (line 91)

obj.layers(l).block.forwardAdvanced(obj.layers(l)) ;

Error in cnn_train_dag_ps>processEpoch (line 250)

net.eval(inputs, params.derOutputs, 'holdOn', s < params.numSubBatches) ;

Error in cnn_train_dag_ps (line 114)

[net, state] = processEpoch(net, state, params, 'train') ;

Error in Dag_Train (line 83)

[ps_net, info] = cnn_train_dag_ps(ps_net, Train_Data, @(i,b) getBatch(bopts,i,b), trainOpts) ;

---

I've not understand that i could resolve this issue, i would be grateful if someone can help me, thanks.

Regards.

Answer 1

Check the structure of net.layers(i).block.size of your network to be compatible with ith input. This compatibility is related to the number of dimensions and also the third component of their dimension.For example, if your input size of the network is MxNx3xP, then the dimension of the first filter should be VxVx3xQ.