SigmoidLayer wont work while implementing on ZC706

Question

Chen 2025-3-3

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2174726-sigmoidlayer-wont-work-while-implementing-on-zc706

回答： Kiran Kintali 2025-3-5

layers = [
    imageInputLayer([800,1,1], 'Name', 'input','Normalization', 'none') % input layers (800, 1, 1)
    % encoder
    convolution2dLayer([3,1], 64, 'Padding', 'same', 'Stride', [1,1], 'Name', 'enc_conv1')
    reluLayer('Name', 'enc_relu1')
    convolution2dLayer([3,1], 32, 'Padding', 'same', 'Stride', [1,1], 'Name', 'enc_conv2')
    reluLayer('Name', 'enc_relu2')
    convolution2dLayer([3,1], 16, 'Padding', 'same', 'Stride', [1,1], 'Name', 'enc_conv3')
    reluLayer('Name', 'enc_relu3')
    convolution2dLayer([3,1], 4, 'Padding', 'same', 'Stride', [1,1], 'Name', 'enc_conv4')
    reluLayer('Name', 'enc_relu4')
    % decoder
    convolution2dLayer([3,1], 16, 'Padding', 'same', 'Stride', [1,1], 'Name', 'dec_conv1')
    reluLayer('Name', 'dec_relu1')
    convolution2dLayer([3,1], 32, 'Padding', 'same', 'Stride', [1,1], 'Name', 'dec_conv2')
    reluLayer('Name', 'dec_relu2')
    convolution2dLayer([3,1], 64, 'Padding', 'same', 'Stride', [1,1], 'Name', 'dec_conv3')
    reluLayer('Name', 'dec_relu3')
    convolution2dLayer([3,1], 1, 'Padding', 'same', 'Stride', [1,1], 'Name', 'dec_output')
    sigmoidLayer('Name', 'sigmoid_output')
    ];
hTarget = dlhdl.Target("Xilinx",Interface="Ethernet",IPAddress="192.168.1.101");
hdlsetuptoolpath('ToolName', 'Xilinx Vivado', 'ToolPath', 'E:\Xilinx\Vivado\2024.2\bin\vivado.bat');
hW = dlhdl.Workflow(Network=net,Bitstream='zc706_single',Target=hTarget);
dn = compile(hW);
deploy(hW);

while I deploy the sigmoidlayer the output from ZC706 fpga the output is zero after sigmoidlayer, and the middle layers output's parameters is correct, it seems that the sigmoidlayers have some problems.How can I fix or it is a bug?