Can you show us the python VGG16 code? Want to improve it for VGG19

[marcodelmoral]

Hello!

Thank you for your project, I love it!

I'm working on a VGG19 network and I want to update your code of VGG16 but I can't find the python code. Can you upload it please?

Kind regards

[chrismaliszewski]

Hey. You haven't posted any code. Have you finished the task?

If not, just for the future reference, here is a piece of code that do most of the job and requires minor changes. I made this code for the model of mine:

#input
    to_input('input.png', width=25.6, height=5.0, name='input'),

    #conv1
    to_ConvConvRelu( name='cr1', s_filer=256, n_filer=(64,64), offset="(0,0,0)", to="(0,0,0)", width=(2,2), height=40, depth=40, caption="conv1"),
    to_Pool(name="p1", offset="(0,0,0)", to="(cr1-east)", width=1, height=35, depth=35, opacity=0.5),

    #conv2
    to_ConvConvRelu( name='cr2', s_filer=128, n_filer=(128,128), offset="(2,0,0)", to="(p1-east)", width=(4,4), height=35, depth=35, caption="conv2"),
    to_Pool(name="p2", offset="(0,0,0)", to="(cr2-east)", width=1, height=30, depth=30, opacity=0.5),

    #conv3
    to_ConvConvRelu( name='cr3', s_filer=64, n_filer=("256","256","256"), offset="(2,0,0)", to="(p2-east)", width=(4,4,4), height=30, depth=30, caption="conv3"),
    to_Pool(name="p3", offset="(0,0,0)", to="(cr3-east)", width=1, height=23, depth=23, opacity=0.5),

    #conv4
    to_ConvConvRelu( name='cr4', s_filer=32, n_filer=("512","512","512"), offset="(2,0,0)", to="(p3-east)", width=(4,4,4), height=23, depth=23, caption="conv4"),
    to_Pool(name="p4", offset="(0,0,0)", to="(cr4-east)", width=1, height=15, depth=15, opacity=0.5),

    #conv5
    to_ConvConvRelu( name='cr5', s_filer=16, n_filer=("512","512","512"), offset="(2,0,0)", to="(p4-east)", width=(4,4,4), height=15, depth=15, caption="conv5"),
    to_Pool(name="p5", offset="(0,0,0)", to="(cr5-east)", width=1, height=10, depth=10, opacity=0.5),

    #flatten
    to_FullyConnected(name="fl", s_filer=4096, offset="(1.25,0,0)", to="(p5-east)", width=1, height=1, depth=20, caption="fl" ),

    #fc1
    to_FullyConnected(name="fc1", s_filer=8192, offset="(1.25,0,0)", to="(fl-east)", width=1, height=1, depth=40, caption="fc1\ndr" ),

    #fc2
    to_FullyConnected(name="fc2", s_filer=8192, offset="(1.25,0,0)", to="(fc1-east)", width=1, height=1, depth=40, caption="fc2\ndr" ),

    #fc3
    to_FullyConnected(name="fc3", s_filer=8192, offset="(1.25,0,0)", to="(fc2-east)", width=1, height=1, depth=40, caption="fc3\ndr" ),

    to_input('output.png', width=25.6, height=5.0, name='output', to='(30, 0, 0)'),

    #sigmoid
    to_SoftMax(name="sg", n_filer="256", offset="(1.25,0,0)", to="(fc3-east)", width=1, height=1, depth=10, caption="SIGMOID", opacity=1.0),

    #connections
    to_connection( "p1", "cr2"),
    to_connection( "p2", "cr3"),
    to_connection( "p3", "cr4"),
    to_connection( "p4", "cr5"),
    to_connection( "p5", "fl"),
    to_connection( "fl", "fc1"),
    to_connection( "fc1", "fc2"),
    to_connection( "fc2", "fc3"),
    to_connection( "fc3", "sg"),
    to_connection( "sg", "++(1.5,0,0)", to_dir=''),

The code above requires minor changes described in these posts: https://github.com/HarisIqbal88/PlotNeuralNet/issues/84 https://github.com/HarisIqbal88/PlotNeuralNet/issues/83 https://github.com/HarisIqbal88/PlotNeuralNet/issues/82 https://github.com/HarisIqbal88/PlotNeuralNet/issues/81

Cheers.