def __init__(self,
dimensionality,
in_planes,
out_planes,
activation=hypothesis.default.activation,
batchnorm=True,
bias=False,
dilation=1,
downsample=None,
stride=1,
groups=1,
width_per_group=64):
super(BasicBlock, self).__init__()
# Load the requested modules depending on the dimensionality.
modules = load_modules(dimensionality)
self.module_convolution = modules[0]
self.module_batchnorm = modules[1]
self.module_maxpool = modules[2]
self.module_adaptive_avg_pool = modules[3]
# Block properties.
self.module_activation = activation
self.activation = activation()
self.bias = bias
self.batchnorm = batchnorm
self.dilation = dilation
self.in_planes = in_planes
self.out_planes = out_planes
self.stride = stride
# Assign the downsampling mapping, if specified.
self.downsample_mapping = downsample
# Build the residual mapping.
self.residual_mapping = self._build_residual_mapping()
def _load_configuration(self, depth):
modules = load_modules(self.dimensionality)
configurations = {
18: load_configuration_18,
34: load_configuration_34,
50: load_configuration_50,
101: load_configuration_101,
152: load_configuration_152
}
# Check if the desired configuration exists.
if depth not in configurations.keys():
raise ValueError("The specified ResNet configuration (", depth,
") does not exist.")
configuration_loader = configurations[depth]
block, blocks_per_layer = configuration_loader()
return block, blocks_per_layer, modules