67cdc47a3477f596844a6eb25a6c5a2b545b824d,keras_/kerascv/models/menet.py,,me_unit,#Any#Any#Any#Any#Any#Any#Any#Any#,88

Before Change

    y = GluonBatchNormalization(name=name + "/s_conv_bn")(y)
    y = nn.Activation("relu", name=name + "/s_conv_activ")(y)

    y = conv1x1(
        out_channels=mid_channels,
        name=name + "/s_evolve_conv")(y)
    y = GluonBatchNormalization(name=name + "/s_evolve_bn")(y)
    y = nn.Activation("sigmoid", name=name + "/s_evolve_activ")(y)

    x = nn.multiply([x, y], name=name + "/mul")

After Change

        x=x,
        in_channels=mid_channels,
        out_channels=side_channels,
        name=name + "/s_merge_conv")
    y = GluonBatchNormalization(name=name + "/s_merge_bn")(y)
    y = nn.Activation("relu", name=name + "/s_merge_activ")(y)

    // depthwise convolution (bottleneck)
    x = depthwise_conv3x3(
        x=x,
        channels=mid_channels,
        strides=(2 if downsample else 1),
        name=name + "/dw_conv2")
    x = GluonBatchNormalization(name=name + "/dw_bn2")(x)

    // evolution
    y = conv3x3(
        x=y,
        in_channels=side_channels,
        out_channels=side_channels,
        strides=(2 if downsample else 1),
        name=name + "/s_conv")
    y = GluonBatchNormalization(name=name + "/s_conv_bn")(y)
    y = nn.Activation("relu", name=name + "/s_conv_activ")(y)

    y = conv1x1(
        x=y,
        in_channels=side_channels,
        out_channels=mid_channels,
        name=name + "/s_evolve_conv")
    y = GluonBatchNormalization(name=name + "/s_evolve_bn")(y)
    y = nn.Activation("sigmoid", name=name + "/s_evolve_activ")(y)

    x = nn.multiply([x, y], name=name + "/mul")

Instances