def __init__(self, pretrained=False):
super().__init__()
# 3*100*100 -> 64*32*32
self.img_conv = nn.Sequential(
conv3x3(3, 64, stride=1, dilation=1),
nn.BatchNorm2d(64),
nn.ReLU(inplace=True),
Bottleneck(64,
64,
stride=2,
dilation=1,
expansion=2,
need_downsample=True),
Bottleneck(64 * 2,
64,
stride=2,
padding=8,
expansion=1,
need_downsample=True),
)
self.features = UNet(in_depth=64)
self.features.fc = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=64, out_features=128, bias=True))
self.visit_model = DPN26()
self.visit_model.linear = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=self.visit_model.linear.in_features,
out_features=64,
bias=True),
)
self.fc = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(192, 9, bias=True),
)
if (pretrained):
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.constant_(m.bias, 0)
def __init__(self, pretrained=False):
super().__init__()
if pretrained:
mdl = pretrainedmodels.__dict__['se_resnext50_32x4d'](
num_classes=1000, pretrained='imagenet')
else:
mdl = pretrainedmodels.__dict__['se_resnext50_32x4d'](
num_classes=1000, pretrained=None)
self.features = list(mdl.children())[:-2]
self.features.append(nn.AdaptiveAvgPool2d(1))
self.features = nn.Sequential(*self.features)
self.features[0].conv1 = nn.Conv2d(3,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.fc_img = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=mdl.last_linear.in_features,
out_features=256,
bias=True),
)
self.visit_model = DPN26()
self.visit_model.linear = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=self.visit_model.linear.in_features,
out_features=128,
bias=True),
)
self.fc = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(384, 9, bias=True),
)
if not (pretrained):
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.constant_(m.bias, 0)
def __init__(self, pretrained=False):
super().__init__()
# print(pretrainedmodels.model_names)
if pretrained:
mdl = pretrainedmodels.__dict__['polynet'](num_classes=1000,
pretrained='imagenet')
else:
mdl = pretrainedmodels.__dict__['polynet'](num_classes=1000,
pretrained=None)
# self.features = list(mdl.children())
self.img_model = mdl
self.img_model.stem.conv1[0].conv = nn.Conv2d(3,
32,
kernel_size=3,
padding=1)
self.img_model.avg_pool = nn.AdaptiveAvgPool2d(1)
self.img_model.last_linear = nn.Linear(
in_features=mdl.last_linear.in_features,
out_features=64,
bias=True)
self.vis_model = DPN26()
self.vis_model.linear = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=self.vis_model.linear.in_features,
out_features=64,
bias=True),
)
self.fc = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(128, 9, bias=True),
)
if not (pretrained):
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.constant_(m.bias, 0)
def __init__(self, pretrained=True):
super().__init__()
if pretrained:
img_model = pretrainedmodels.__dict__['se_resnext101_32x4d'](
num_classes=1000, pretrained='imagenet') #seresnext101
else:
img_model = pretrainedmodels.__dict__['se_resnext101_32x4d'](
num_classes=1000, pretrained=None)
self.visit_model = DPN26()
self.img_encoder = list(img_model.children())[:-2]
self.img_encoder.append(nn.AdaptiveAvgPool2d(1))
self.img_encoder = nn.Sequential(*self.img_encoder)
self.img_fc = nn.Sequential(
FCViewer(), nn.Dropout(0.5),
nn.Linear(img_model.last_linear.in_features, 256))
self.cls = nn.Linear(320, 9)
def __init__(self, pretrained=False):
super().__init__()
self.visit_model = DPN26()
self.cls = nn.Linear(64, 9)