def __init__(self,
channels,
reduction=16,
act_layer=nn.ReLU,
divisor=1,
reduction_channels=None,
gate_layer='sigmoid'):
super(SEWithNorm, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
reduction_channels = reduction_channels or make_divisible(
channels // reduction, divisor=divisor)
print(reduction_channels)
self.fc1 = nn.Conv2d(channels,
reduction_channels,
kernel_size=1,
padding=0,
bias=True)
self.bn = nn.BatchNorm2d(reduction_channels)
self.act = act_layer(inplace=True)
self.fc2 = nn.Conv2d(reduction_channels,
channels,
kernel_size=1,
padding=0,
bias=True)
self.gate = create_act_layer(gate_layer)
def _run(x, act_layer=''):
if act_layer:
# replace act layer if set
m.act = create_act_layer(act_layer, inplace=inplace)
out = m(x)
l = (out - 0).pow(2).sum()
return l
def __init__(self,
in_channels,
out_channels,
kernel_size,
padding=0,
stride=1,
use_batchnorm=True,
act_layer='relu'):
conv = nn.Conv2d(
in_channels,
out_channels,
kernel_size,
stride=stride,
padding=padding,
bias=not (use_batchnorm),
)
if act_layer == 'frelu':
act = FReLU(out_channels)
elif act_layer == 'dy_relu_b':
act = DyReLUB(out_channels)
elif act_layer == 'dy_relu_c':
act = DyReLUC(out_channels)
else:
act = create_act_layer(act_layer)
if use_batchnorm:
bn = nn.BatchNorm2d(out_channels)
else:
bn = nn.Identity()
super(Conv2dAct, self).__init__(conv, bn, act)
def __init__(self, act_layer="relu", inplace=True):
super(MLP, self).__init__()
self.fc1 = nn.Linear(1000, 100)
self.act = create_act_layer(act_layer, inplace=inplace)
self.fc2 = nn.Linear(100, 10)