public ConvNet(ConvNetArgs args)
: base(args)
{
var layers = keras.layers;
// Convolution Layer with 32 filters and a kernel size of 5.
conv1 = layers.Conv2D(32, kernel_size: 5, activation: keras.activations.Relu);
// Max Pooling (down-sampling) with kernel size of 2 and strides of 2.
maxpool1 = layers.MaxPooling2D(2, strides: 2);
// Convolution Layer with 64 filters and a kernel size of 3.
conv2 = layers.Conv2D(64, kernel_size: 3, activation: keras.activations.Relu);
// Max Pooling (down-sampling) with kernel size of 2 and strides of 2.
maxpool2 = layers.MaxPooling2D(2, strides: 2);
// Flatten the data to a 1-D vector for the fully connected layer.
flatten = layers.Flatten();
// Fully connected layer.
fc1 = layers.Dense(1024);
// Apply Dropout (if is_training is False, dropout is not applied).
dropout = layers.Dropout(rate: 0.5f);
// Output layer, class prediction.
output = layers.Dense(args.NumClasses);
StackLayers(conv1, maxpool1, conv2, maxpool2, flatten, fc1, dropout, output);
}
public ConvNet(ConvNetArgs args)
: base(args)
{
// Convolution Layer with 32 filters and a kernel size of 5.
conv1 = Conv2D(32, kernel_size: 5, activation: tf.nn.relu);
// Max Pooling (down-sampling) with kernel size of 2 and strides of 2.
maxpool1 = MaxPooling2D(2, strides: 2);
// Convolution Layer with 64 filters and a kernel size of 3.
conv2 = Conv2D(64, kernel_size: 3, activation: tf.nn.relu);
// Max Pooling (down-sampling) with kernel size of 2 and strides of 2.
maxpool2 = MaxPooling2D(2, strides: 2);
// Flatten the data to a 1-D vector for the fully connected layer.
flatten = Flatten();
// Fully connected layer.
fc1 = Dense(1024);
// Apply Dropout (if is_training is False, dropout is not applied).
dropout = Dropout(rate: 0.5f);
// Output layer, class prediction.
output = Dense(args.NumClasses);
}
