def __init__(self,
depth=18,
init_plane=64,
out_plane=None,
stage=4,
num_class=10,
small_input=True,
doublechannel=None,
downsample=None):
"""Create layers.
:param num_reps: number of layers
:type num_reqs: int
:param items: channel and stride of every layer
:type items: dict
:param num_class: number of class
:type num_class: int
"""
super(ResNet, self).__init__()
self.backbone = ResNetGeneral(small_input, init_plane, depth, stage,
doublechannel, downsample)
self.adaptiveAvgPool2d = AdaptiveAvgPool2d(output_size=(1, 1))
self.view = View()
out_plane = out_plane or self.backbone.output_channel
self.head = Linear(in_features=out_plane, out_features=num_class)
def __init__(self, **descript):
"""Create layers."""
super().__init__()
# state_dim = descript.get("state_dim")
action_dim = descript.get("action_dim")
self.lambda1 = Lambda(lambda x: tf.cast(x, dtype='float32') / 255.)
self.conv1 = Conv2d(in_channels=4,
out_channels=32,
kernel_size=8,
stride=4,
bias=False)
self.ac1 = Relu()
self.conv2 = Conv2d(in_channels=32,
out_channels=64,
kernel_size=4,
stride=2,
bias=False)
self.ac2 = Relu()
self.conv3 = Conv2d(in_channels=64,
out_channels=64,
kernel_size=3,
stride=1,
bias=False)
self.ac3 = Relu()
self.view = View()
self.fc1 = Linear(64, 256)
self.ac4 = Relu()
self.fc2 = Linear(256, action_dim)
def __init__(self,
code=None,
depth=18,
base_channel=64,
out_plane=2048,
stage=4,
num_class=1000,
small_input=True,
block='BasicBlock',
pretrained_arch=None,
pretrained=None):
"""Create layers.
:param num_reps: number of layers
:type num_reqs: int
:param items: channel and stride of every layer
:type items: dict
:param num_class: number of class
:type num_class: int
"""
super(FasterBackbone, self).__init__()
self.backbone = SpResNetDet(depth=depth,
block=block,
code=code,
pretrained=pretrained,
pretrained_arch=pretrained_arch)
self.adaptiveAvgPool2d = AdaptiveAvgPool2d(output_size=(1, 1))
self.view = View()
out_plane = out_plane or self.backbone.out_channels
self.head = Linear(in_features=out_plane, out_features=num_class)