Error: A Layer Sequence with 'nnet.cnn.​layer.Batc​hNormaliza​tionLayer' immediately following 'nnet.cnn.​layer.ReLU​Layer' is not currently supported.

Hello,

I have a pretrained NN imported from TensorFlow in a DAGNetwork object. However, when I try to compile the NN with Deep Learning HDL Toolbox, I get the following error:

Error using dnnfpga.compiler.codegenfpga
A Layer Sequence with  'nnet.cnn.layer.BatchNormalizationLayer' immediately following  'nnet.cnn.layer.ReLULayer' is not currently
supported. A Batchnorm layer must follow a convolutional layer.

The NN has not the sequence BatchNormalization layer + ReLu layer. BatchNormalization layer is always followed by a convolution layer as it is suggested.

Find the NN in the following link: NN.mat

Could you help me with this issue?

Answer 1

[Donald Baltus]

Hello,

The error message says the network cannot include a batch normalization layer immediately AFTER a relu l;ayer. This newtork includes that sequence in several places (for instance layers 5 and 6).

Please follow up if you have more questions.