def __init__(self, params):
"""
:param params: {'num_channels':1,
'num_filters':64,
'kernel_h':5,
'kernel_w':5,
'stride_conv':1,
'pool':2,
'stride_pool':2,
'num_classes':28
'se_block': False,
'drop_out':0.2}
"""
super(QuickNat, self).__init__()
self.encode1 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode3 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode4 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.bottleneck = sm.DenseBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode2 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode3 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode4 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
def __init__(self, params):
"""
:param params: {'num_channels':1,
'num_filters':64,
'kernel_h':5,
'kernel_w':5,
'stride_conv':1,
'pool':2,
'stride_pool':2,
'num_classes':28
'se_block': False,
'drop_out':0.2}
"""
super(QuickFCNClassifier, self).__init__()
self.encode1 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode3 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode4 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.bottleneck = sm.DenseBlock(params, se_block_type=se.SELayer.CSSE)
############Classification Task############
self.classifier = nn.Sequential(nn.Linear(40000, 25), nn.PReLU(),
nn.Linear(25, 3))
def __init__(self, params):
"""
:param params: {'num_channels':1,
'num_filters':64,
'kernel_h':5,
'kernel_w':5,
'stride_conv':1,
'pool':2,
'stride_pool':2,
'num_classes':28
'se_block': False,
'drop_out':0.2}
"""
super(QuickResNet, self).__init__()
self.resultnet = ResUltNet(
params) # TODO: Is it right to pass params here too?
self.unbottle = nn.Sequential(
# input is Z, going into a convolution
nn.ConvTranspose2d(64, 32, 4, 1, 0, bias=False),
nn.BatchNorm2d(32),
nn.ReLU(True),
# state size. (ngf*8) x 4 x 4
nn.ConvTranspose2d(32, 16, 4, 2, 1, bias=False),
nn.BatchNorm2d(16),
nn.ReLU(True),
# state size. (ngf*4) x 8 x 8
nn.ConvTranspose2d(16, 8, 4, 2, 1, bias=False),
nn.BatchNorm2d(8),
nn.ReLU(True),
# state size. (ngf*2) x 16 x 16
nn.ConvTranspose2d(8, 3, 4, 2, 1, bias=False),
nn.BatchNorm2d(3),
nn.ReLU(True))
self.encode1 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode3 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode4 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.bottleneck = sm.DenseBlock(params, se_block_type=se.SELayer.CSSE)
self.conv1 = torch.nn.Conv2d(params['num_channels'], 3, kernel_size=1)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode2 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode3 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode4 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
def __init__(self, params):
super(Segmentor, self).__init__()
params['num_channels'] = 1
self.encode1 = sm.EncoderBlock(params)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params)
self.encode3 = sm.EncoderBlock(params)
self.bottleneck = sm.DenseBlock(params)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params, se_block_type=se.SELayer.NONE)
self.decode2 = sm.DecoderBlock(params, se_block_type=se.SELayer.NONE)
self.decode3 = sm.DecoderBlock(params, se_block_type=se.SELayer.NONE)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
self.sigmoid = nn.Sigmoid()
def __init__(self, params):
super(SDnetSegmentor, self).__init__()
params['num_channels'] = 1
params['num_filters'] = 64
self.encode1 = sm.EncoderBlock(params)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params)
self.encode3 = sm.EncoderBlock(params)
self.bottleneck = sm.GenericBlock(params)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params)
self.decode2 = sm.DecoderBlock(params)
self.decode3 = sm.DecoderBlock(params)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
self.soft_max = nn.Softmax2d()
def __init__(self, params):
super(QuickNat, self).__init__()
self.encode1 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode3 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode4 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.bottleneck = sm.DenseBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode2 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode3 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode4 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
def __init__(self):
super(Network, self).__init__()
params['num_channels'] = 1
params['num_class'] = pretrained_num_classes
self.encode1 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.encode2 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode3 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.encode4 = sm.EncoderBlock(params, se_block_type=se.SELayer.CSSE)
self.bottleneck = sm.DenseBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 128
self.decode1 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode2 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode3 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
self.decode4 = sm.DecoderBlock(params, se_block_type=se.SELayer.CSSE)
params['num_channels'] = 64
self.classifier = sm.ClassifierBlock(params)
def __init__(self, params):
"""
:param params: {'num_channels':1,
'num_filters':64,
'kernel_h':5,
'kernel_w':5,
'stride_conv':1,
'pool':2,
'stride_pool':2,
'num_classes':28
'se_block': False,
'drop_out':0.2}
"""
super(QuickOct, self).__init__()
print("NUMBER OF CHANNEL", params['num_channels'])
self.encode1 = sm.EncoderBlock(params,
se_block_type=params['se_block'])
params['num_channels'] = params['num_filters']
self.encode2 = sm.OctaveEncoderBlock(params,
se_block_type=params['se_block'])
self.encode3 = sm.OctaveEncoderBlock(params,
se_block_type=params['se_block'])
# self.encode4 = sm.OctaveEncoderBlock(params, se_block_type=params['se_block'])
self.bottleneck = sm.OctaveDenseBlock(params,
se_block_type=params['se_block'])
params['num_channels'] = params['num_filters'] * 2
self.decode1 = sm.OctaveDecoderBlock(params,
se_block_type=params['se_block'])
self.decode2 = sm.OctaveDecoderBlock(params,
se_block_type=params['se_block'])
self.decode3 = sm.DecoderBlock(params,
se_block_type=params['se_block'])
# self.decode4 = sm.DecoderBlock(params, se_block_type=params['se_block'])
params['num_channels'] = params['num_filters']
self.classifier = sm.ClassifierBlock(params)