def __init__(self,
in_dim,
projectionFactor,
out_dim,
dilation,
dtype=torch.float32):
super(BottleNeckDownSamplingDilatedConvLast, self).__init__()
# Main branch
# Secondary branch
self.block0 = conv_block_1(in_dim, int(in_dim / projectionFactor))
self.conv1 = nn.Conv2d(int(in_dim / projectionFactor),
int(in_dim / projectionFactor),
kernel_size=3,
padding=dilation,
dilation=dilation)
self.bn1 = nn.BatchNorm2d(int(in_dim / projectionFactor))
self.PReLU1 = nn.PReLU()
self.block2 = conv_block_1(int(in_dim / projectionFactor), out_dim)
self.do = nn.Dropout(p=0.01)
self.conv_out = nn.Conv2d(in_dim, out_dim, kernel_size=3, padding=1)
self.PReLU3 = nn.PReLU()
def __init__(self, in_dim, projectionFactor, out_dim, dtype=torch.float32):
super().__init__()
self.dtype = dtype
# Main branch
self.maxpool0 = nn.MaxPool2d(2, return_indices=True)
# Secondary branch
self.conv0 = nn.Conv2d(in_dim,
int(in_dim / projectionFactor),
kernel_size=2,
stride=2)
self.bn0 = nn.BatchNorm2d(int(in_dim / projectionFactor))
self.PReLU0 = nn.PReLU()
self.conv1 = nn.Conv2d(int(in_dim / projectionFactor),
int(in_dim / projectionFactor),
kernel_size=3,
padding=1)
self.bn1 = nn.BatchNorm2d(int(in_dim / projectionFactor))
self.PReLU1 = nn.PReLU()
self.block2 = conv_block_1(int(in_dim / projectionFactor), out_dim)
self.do = nn.Dropout(p=0.01)
self.PReLU3 = nn.PReLU()
def __init__(self, in_dim, out_dim, projectionFactor, dropoutRate):
super(BottleNeckNormal_Asym, self).__init__()
self.in_dim = in_dim
self.out_dim = out_dim
# Main branch
# Secondary branch
self.block0 = conv_block_1(in_dim, int(in_dim / projectionFactor))
self.block1 = conv_block_Asym(int(in_dim / projectionFactor), int(in_dim / projectionFactor), 5)
self.block2 = conv_block_1(int(in_dim / projectionFactor), out_dim)
self.do = nn.Dropout(p=dropoutRate)
self.PReLU_out = nn.PReLU()
if in_dim > out_dim:
self.conv_out = conv_block_1(in_dim, out_dim)
def __init__(self, in_dim, projectionFactor, out_dim):
super(BottleNeckUpSampling, self).__init__()
# Main branch
self.conv0 = nn.Conv2d(in_dim, int(in_dim / projectionFactor), kernel_size=3, padding=1)
self.bn0 = nn.BatchNorm2d(int(in_dim / projectionFactor))
self.PReLU0 = nn.PReLU()
self.conv1 = nn.Conv2d(int(in_dim / projectionFactor), int(in_dim / projectionFactor), kernel_size=3, padding=1)
self.bn1 = nn.BatchNorm2d(int(in_dim / projectionFactor))
self.PReLU1 = nn.PReLU()
self.block2 = conv_block_1(int(in_dim / projectionFactor), out_dim)
self.do = nn.Dropout(p=0.01)
self.PReLU3 = nn.PReLU()
