def __init__(self,
input_shape,
fc=True,
num_classes=1000,
first_output=16,
growth_rate=12,
num_blocks=3,
depth=40,
dropout=False,
name='DenseNet'):
super(DenseNet, self).__init__(input_shape=input_shape,
layer_name=name)
self.append(
nn.ConvolutionalLayer(self.input_shape,
first_output,
3,
activation='linear',
layer_name=name + 'pre_conv'))
n = (depth - 1) // num_blocks
for b in range(num_blocks):
self.append(
nn.DenseBlock(self.output_shape,
num_conv_layer=n - 1,
growth_rate=growth_rate,
dropout=dropout,
layer_name=name + 'dense_block_%d' % b))
if b < num_blocks - 1:
self.append(
nn.DenseBlock(self.output_shape,
True,
None,
None,
dropout,
layer_name=name +
'dense_block_transit_%d' % b))
self.append(
nn.BatchNormLayer(self.output_shape, layer_name=name + 'post_bn'))
if fc:
self.append(
nn.GlobalAveragePoolingLayer(input_shape,
name + '_glbavgpooling'))
self.append(
nn.SoftmaxLayer(self.output_shape, num_classes,
name + '_softmax'))
def __init__(self,
input_shape,
block,
layers,
num_filters,
activation='relu',
fc=True,
pooling=True,
num_classes=1000,
layer_name='ResNet',
**kwargs):
super(ResNet, self).__init__(input_shape=input_shape,
layer_name=layer_name)
self.activation = activation
self.custom_block = kwargs.pop('custom_block', None)
self.kwargs = kwargs
self.append(
nn.ConvNormAct(self.input_shape,
num_filters,
7,
stride=2,
activation=activation,
**kwargs))
if pooling:
self.append(
nn.PoolingLayer(self.output_shape, (3, 3),
stride=(2, 2),
pad=1))
self.append(
self._make_layer(block,
self.output_shape,
num_filters,
layers[0],
name='block1'))
self.append(
self._make_layer(block,
self.output_shape,
2 * num_filters,
layers[1],
stride=2,
name='block2'))
self.append(
self._make_layer(block,
self.output_shape,
4 * num_filters,
layers[2],
stride=2,
name='block3'))
self.append(
self._make_layer(block,
self.output_shape,
8 * num_filters,
layers[3],
stride=2,
name='block4'))
if fc:
self.append(
nn.GlobalAveragePoolingLayer(self.output_shape,
layer_name='glb_avg_pooling'))
self.append(
nn.FullyConnectedLayer(self.output_shape,
num_classes,
activation='softmax',
layer_name='output'))