def __init__(_,net_name,out_channels):
super().__init__()
_.net_name = net_name
_.shapes_have_been_printed = False
_.a = nn.Conv2d(a, b, kernel_size=3, stride=2)
_.b = nn.ReLU(inplace=True)
_.c = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.d = Fire(b, b_, c, c)
_.e = Fire(c+c, b, d, d)
_.f = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.g = Fire(e, c, e, e)
_.i = Fire(e+e, c, e, e)
_.j = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.k = nn.Upsample((32,32),mode='nearest')
_.k2 = nn.Upsample((64,64),mode='nearest')
_.k3 = nn.Upsample((200,200),mode='nearest')
_.l = nn.Conv2d(
in_channels=d+d,
out_channels=out_channels,
padding=1,
kernel_size=3)
_.l2 = nn.Conv2d(
in_channels=e+e,
out_channels=out_channels,
padding=1,
kernel_size=3)
_.o = nn.AvgPool2d(2, stride=2)
_.bn3 = nn.BatchNorm2d(a)
_.bn32 = nn.BatchNorm2d(d)
_.bn64 = nn.BatchNorm2d(e)
def __init__(self,net_name):
super().__init__()
self.net_name = net_name
self.shapes_have_been_printed = True
self.a = nn.Conv2d(a, d, kernel_size=3, stride=2)
self.b = nn.ReLU(inplace=True)
self.c = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
self.d = Fire(d, b, d, d)
self.e = Fire(d+d, b, d, d)
self.f = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
self.g = Fire(e, c, e, e)
self.i = Fire(e+e, c, e, e)
self.j = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
self.k = nn.Upsample((32,32),mode='nearest')
self.k2 = nn.Upsample((64,64),mode='nearest')
self.k3 = nn.Upsample((140,140),mode='nearest')
self.l = nn.Conv2d(
in_channels=d+d,
out_channels=a,
padding=1,
kernel_size=3)
self.l2 = nn.Conv2d(
in_channels=e+e,
out_channels=a,
padding=1,
kernel_size=3)
self.o = nn.AvgPool2d(2, stride=2)
self.bn3 = nn.BatchNorm2d(3)
self.bn32 = nn.BatchNorm2d(32)
self.bn64 = nn.BatchNorm2d(64)
def __init__(_, net_name, out_channels):
super().__init__()
_.A = {}
_.quant = torch.quantization.QuantStub()
_.dequant = torch.quantization.DeQuantStub()
_.net_name = net_name
_.shapes_have_been_printed = False
_.a = torch.nn.Conv2d(a, b_, kernel_size=3, stride=2)
_.b = nn.ReLU(inplace=True)
_.c = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.d = Fire(b_, b_, b_, b_)
_.e = Fire(b, b, b_, b_)
_.f = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.g = Fire(e, c, c, c)
_.i = Fire(c + c, c, c, c)
_.j = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.k = nn.Upsample((32, 32), mode='nearest')
_.k2 = nn.Upsample((64, 64), mode='nearest')
_.k200 = nn.Upsample((200, 200), mode='nearest')
_.k1000 = nn.Upsample((1000, 1000), mode='nearest')
_.l = nn.Conv2d(in_channels=d + d,
out_channels=out_channels,
padding=1,
kernel_size=3)
_.l2 = nn.Conv2d(in_channels=b,
out_channels=out_channels,
padding=0,
kernel_size=1)
_.o = nn.AvgPool2d(2, stride=2)
_.bn3 = nn.Identity() #nn.BatchNorm2d(a)
_.bn32 = nn.Identity() #nn.BatchNorm2d(d)
_.bn64 = nn.Identity() #nn.BatchNorm2d(e)
def __init__(_, net_name, in_channels, out_channels):
super().__init__()
_.quant = torch.quantization.QuantStub()
_.dequant = torch.quantization.DeQuantStub()
_.net_name = net_name
_.shapes_have_been_printed = False
_.a = nn.Conv2d(in_channels, d, kernel_size=3, stride=2)
_.b = nn.ReLU(inplace=True)
_.c = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.d0 = Fire(35, 32, 32, 32)
_.d0_ = Fire(38, 32, 32, 32)
_.d1 = Fire(40, 32, 32, 32)
_.d = Fire(d, b, d, d)
_.e = Fire(d + d, b, b, b)
_.f = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.g = Fire(e, c, e, e)
_.i = Fire(e + e, c, e, e)
_.j = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
_.k = nn.Upsample((32, 32), mode='nearest')
_.k2 = nn.Upsample((64, 64), mode='nearest')
_.k200 = nn.Upsample((200, 200), mode='nearest')
_.k1000 = nn.Upsample((1000, 1000), mode='nearest')
_.l = nn.Conv2d(in_channels=d + d,
out_channels=out_channels,
padding=1,
kernel_size=3)
_.l2 = nn.Conv2d(in_channels=e + e,
out_channels=out_channels,
padding=1,
kernel_size=6)
_.o = nn.AvgPool2d(2, stride=2)
_.bn3 = nn.BatchNorm2d(a)
_.bn32 = nn.BatchNorm2d(d)
_.bn64 = nn.BatchNorm2d(e)