c2041cf9be09dfbc5f975ab21fa1d318ad0f2c24,batchflow/models/torch/modules.py,FPA,__init__,#FPA#Any#Any#Any#Any#Any#Any#Any#Any#,177

Before Change

        branches = [dict(layout=layout, filters="same", kernel_size=1)] // the mid branch from inputs tensor directly

        if use_dilation:
            base_kernel_size = pyramid_kernel_size[-1] // 3
            main_kernel_size = [base_kernel_size] * (depth + 1) + [factor] * (depth) // 3 3 3 3 2 2 2
            // infering coresponding dilation for every kernel_size in pyramid block
            pyramid_dilation = [round((rf - base_kernel_size) / (base_kernel_size - 1)) + 1

After Change

        if bottleneck:
            bottleneck = 4 if bottleneck is True else bottleneck
            out_filters = get_num_channels(inputs)
            inner_filters = out_filters // bottleneck
            self.pyramid = ConvBlock(dict(layout=layout, kernel_size=1, filters=inner_filters),
                                     pyramid_args,
                                     dict(layout=layout, kernel_size=1, filters=out_filters), inputs=inputs, **kwargs)

Instances