def __init__(self, v_compress, shrink, bits, fuse_level, level):
super(DecoderCell, self).__init__()
# Init.
self.v_compress = v_compress
self.fuse_level = fuse_level
print('\tDecoder fuse level: {}'.format(self.fuse_level))
# Layers.
self.conv1 = nn.Conv2d(bits * (level + 1),
512,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.rnn1 = ConvLSTMCell(512,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(
(((128 + 256 // shrink * 2) if v_compress else 128)
if self.fuse_level >= 3 else 128), #out1=256
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(
(((128 + 128 // shrink * 2) if v_compress else 128)
if self.fuse_level >= 2 else 128), #out2=128
256,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.rnn4 = ConvLSTMCell(
(64 + 64 // shrink * 2) if v_compress else 64, #out3=64
128,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.conv2 = nn.Conv2d(32,
3,
kernel_size=1,
stride=1,
padding=0,
bias=False)
def __init__(self):
super(EncoderCell, self).__init__()
self.conv = nn.Conv2d(
3, 64, kernel_size=3, stride=2, padding=1, bias=False)
self.rnn1 = ConvLSTMCell(
64,
256,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(
256,
512,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(
512,
512,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
def __init__(self, v_compress, stack, fuse_encoder, fuse_level):
super(EncoderCell, self).__init__()
# Init.
self.v_compress = v_compress
self.fuse_encoder = fuse_encoder
self.fuse_level = fuse_level
if fuse_encoder:
print('\tEncoder fuse level: {}'.format(self.fuse_level))
# Layers.
self.conv = nn.Conv2d(9 if stack else 3,
64,
kernel_size=3,
stride=2,
padding=1,
bias=False)
self.rnn1 = ConvLSTMCell(
64, #128 if fuse_encoder and v_compress else 64,
128,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(
128, #((384 if fuse_encoder and v_compress else 256)
#if self.fuse_level >= 2 else 256),
128,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(
128, #((384 if fuse_encoder and v_compress else 512)
#if self.fuse_level >= 3 else 256),
128,
kernel_size=4,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
def __init__(self, v_compress, stack, fuse_encoder, fuse_level):
super(EncoderCell, self).__init__()
# Init.
self.v_compress = v_compress
self.fuse_encoder = fuse_encoder
self.fuse_level = fuse_level
if fuse_encoder:
print('\tEncoder fuse level: {}'.format(self.fuse_level))
# Layers.
self.conv = nn.Conv2d(9 if stack else 3,
64,
kernel_size=3,
stride=2,
padding=1,
bias=False)
for param in self.conv.parameters():
param.requires_grad = False
self.rnn1 = ConvLSTMCell(128 if fuse_encoder and v_compress else 64,
256,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
# print(fuse_encoder,"fuse_encoder",v_compress,"v_compress")
self.rnn2 = ConvLSTMCell(((384 if fuse_encoder and v_compress else 256)
if self.fuse_level >= 2 else 256),
512,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(((768 if fuse_encoder and v_compress else 512)
if self.fuse_level >= 3 else 512),
512,
kernel_size=3,
stride=2,
padding=1,
hidden_kernel_size=1,
bias=False)
def __init__(self, bottleneck, output_channels):
super(DecoderCell, self).__init__()
#self.conv1= nn.Conv2d(16, 32, kernel_size=1, stride=1, padding=0, bias=False)
self.conv1 = nn.Conv2d(bottleneck,
512,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.rnn1 = ConvLSTMCell(512,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(128,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(128,
256,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.rnn4 = ConvLSTMCell(64,
128,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.conv2 = nn.Conv2d(32,
output_channels,
kernel_size=1,
stride=1,
padding=0,
bias=False)
def __init__(self):
super(DecoderCell, self).__init__()
self.conv1 = nn.Conv2d(32,
512,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.rnn1 = ConvLSTMCell(512,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(128,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(128,
256,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.rnn4 = ConvLSTMCell(64,
128,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.conv2 = nn.Conv2d(32,
3,
kernel_size=1,
stride=1,
padding=0,
bias=False)
def __init__(self):
super(DecoderCell, self).__init__()
self.conv1 = nn.Conv2d(
32, 512, kernel_size=1, stride=1, padding=0, bias=False) # changed input channels from 32 to 16 to accomodate the above change
self.rnn1 = ConvLSTMCell(
512,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(
128,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(
128,
256,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.rnn4 = ConvLSTMCell(
64,
128,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.conv2 = nn.Conv2d(
32, 3, kernel_size=1, stride=1, padding=0, bias=False)
def __init__(self, v_compress, shrink, bits, fuse_level):
super(DecoderCell2, self).__init__()
# Init.
self.v_compress = v_compress
self.fuse_level = fuse_level
print('\tDecoder fuse level: {}'.format(self.fuse_level))
# Layers.
#self.conv1 = nn.Conv2d(
# bits*10, 512, kernel_size=1, stride=1, padding=0, bias=False)
self.conv1 = nn.Conv2d(bits*10, 64*10, groups=10, kernel_size=1, stride=1, padding=0, bias=False) # groups=10
self.conv2 = nn.Conv2d(64*10, 128*10, groups=10, kernel_size=1, stride=1, padding=0, bias=False)
self.conv3 = nn.Conv2d(128*10, 512*10, groups=10, kernel_size=1, stride=1, padding=0, bias=False)
#self.bn1 = nn.BatchNorm2d(64*10)
#self.bn2 = nn.BatchNorm2d(128*10)
#self.bn3 = nn.BatchNorm2d(512*10)
self.rnn1 = ConvLSTMCell(
512,
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn2 = ConvLSTMCell(
(((128 + 256 // shrink * 2) if v_compress else 128)
if self.fuse_level >= 3 else 128), #out1=256
512,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=1,
bias=False)
self.rnn3 = ConvLSTMCell(
(((128 + 128//shrink*2) if v_compress else 128)
if self.fuse_level >= 2 else 128), #out2=128
256,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.rnn4 = ConvLSTMCell(
(64 + 64//shrink*2) if v_compress else 64, #out3=64
128,
kernel_size=3,
stride=1,
padding=1,
hidden_kernel_size=3,
bias=False)
self.conv_end = nn.Conv2d(
32,
3,
kernel_size=1, stride=1, padding=0, bias=False)
