class ConvBlock(DeepModel):
    __scope__ = "model.convblock"

    with Settings.default.scope(__scope__) as hparams:
        hparams.kernel_size = (3, 3)
        hparams.dropout = 0.25
        hparams.batch_norm = True
        hparams.stride = 1
        hparams.activation = "relu"

    def __init__(self, **kwargs):
        super(ConvBlock, self).__init__(**kwargs)

        with self.scope() as p:
            # setup default value
            if p.filters is None:
                p.filters = p.in_channels

            layers = []
            # build standard convolution block
            p.padding = [dim_size // 2 for dim_size in p.kernel_size]
            layers.append(
                nn.Conv2d(
                    p.in_channels, p.filters, p.kernel_size, p.stride, padding=p.padding
                )
            )
            if p.batch_norm:
                layers.append(nn.BatchNorm2d(p.filters))
            if p.activation != False:
                layers.append(Activation(p.activation))
            if p.dropout > 0:
                layers.append(nn.Dropout2d(p.dropout))

            # return sequential network
            self.conv = nn.Sequential(*layers)

    def output_shape(self, input_shape: tuple) -> tuple:
        assert type(input_shape) == tuple

        h_in, w_in = input_shape
        with self.scope() as p:
            h_out = (h_in + 2 * p.padding[0] - p.kernel_size[0]) // p.stride + 1
            w_out = (w_in + 2 * p.padding[1] - p.kernel_size[1]) // p.stride + 1
        return (h_out, w_out)

    def forward(self, x):
        return self.conv(x)