Function block_inception_b

tests/utils/custom_layers/inception_blocks.py:103–166

Builds Inception-B block for Inception v4 network.

(inputs, scope=None, is_train=False)

def block_inception_b(inputs, scope=None, is_train=False):
    """Builds Inception-B block for Inception v4 network."""
    # By default use stride=1 and SAME padding

    with tf.variable_scope(scope, 'BlockInceptionB', [inputs]):
        with tf.variable_scope('Branch_0'):
            branch_0, _ = conv_module(
                inputs, n_out_channel=384, filter_size=(1, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0a_1x1'
            )

        with tf.variable_scope('Branch_1'):
            branch_1, _ = conv_module(
                inputs, n_out_channel=192, filter_size=(1, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0a_1x1'
            )

            branch_1, _ = conv_module(
                branch_1, n_out_channel=224, filter_size=(1, 7), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0b_1x7'
            )

            branch_1, _ = conv_module(
                branch_1, n_out_channel=256, filter_size=(7, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0c_7x1'
            )

        with tf.variable_scope('Branch_2'):
            branch_2, _ = conv_module(
                inputs, n_out_channel=192, filter_size=(1, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0a_1x1'
            )

            branch_2, _ = conv_module(
                branch_2, n_out_channel=192, filter_size=(7, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0b_7x1'
            )

            branch_2, _ = conv_module(
                branch_2, n_out_channel=224, filter_size=(1, 7), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0c_1x7'
            )

            branch_2, _ = conv_module(
                branch_2, n_out_channel=224, filter_size=(7, 1), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0d_7x1'
            )

            branch_2, _ = conv_module(
                branch_2, n_out_channel=256, filter_size=(1, 7), strides=(1, 1), padding='SAME', batch_norm_init=None,
                is_train=is_train, use_batchnorm=True, activation_fn='ReLU', name='Conv2d_0e_1x7'
            )

        with tf.variable_scope('Branch_3'):
            branch_3 = tl.layers.MeanPool2d(
                inputs, filter_size=(3, 3), strides=(1, 1), padding='SAME', name='AvgPool_0a_3x3'
            )