class MultiConv(Layer):
def __init__(self,
convs,
channelsAxis=-1,
trainable=True,
name=None,
dtype=None):
super().__init__(trainable, name, dtype)
self.convs = convs
self.concat = Concatenate(channelsAxis)
def build(self, input_shape):
for conv in self.convs:
conv.build(input_shape)
self.concat.build(
[conv.compute_output_shape(input_shape) for conv in self.convs])
super().build(input_shape)
def call(self, inputs, **kwargs):
appliedConvs = []
for conv in self.convs:
appliedConvs.append(conv.call(inputs))
return self.concat.call(appliedConvs)
