def __init__(self, depth_list=(16, 32, 64), num_classes=10, init=None):
super(ResNextMini, self).__init__(depth_list, num_classes, init)
self.block1 = self._build_block(ResNextBlock, depth_list[0],
depth_list[0])
self.block2 = self._build_block(ResNextBlock, depth_list[0],
depth_list[1], 2)
self.block3 = self._build_block(ResNextBlock, depth_list[1],
depth_list[2], 2)
apply_init(init, self.modules())
def __init__(self, depth_list=(16, 32, 64), num_classes=10, init=None):
super(ResNetMini, self).__init__()
self.conv1 = nn.Conv2d(1, depth_list[0], 3, 1, 1)
self.bn1 = nn.BatchNorm2d(num_features=depth_list[0])
self.relu = nn.ReLU(inplace=True)
self.block1 = self._build_block(ResBlock, depth_list[0], depth_list[0])
self.block2 = self._build_block(ResBlock, depth_list[0], depth_list[1],
2)
self.block3 = self._build_block(ResBlock, depth_list[1], depth_list[2],
2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(depth_list[2], num_classes)
# apply the weights initiator
apply_init(init, self.modules())
def __init__(self, pretrained=False, num_classes=10, init=None):
super(ResNet18, self).__init__()
self.model = resnet18(pretrained, num_classes=num_classes)
# apply the weights initiator
apply_init(init, self.model.modules())
def __init__(self, num_classes=10, init=None):
super(DenseNetMini, self).__init__(
growth_rate=12, block_config=(3, 6, 12, 8),
num_init_features=64, num_classes=num_classes
)
apply_init(init, self.modules())