def classification_model_dpn92(**kwargs):
base_model = pretrainedmodels.dpn92()
return ClassificationModelDPN(base_model,
base_model_features=2688,
nb_features=6,
base_model_l1_outputs=64,
**kwargs)
def __init__(self):
self.model = dpn92(pretrained='imagenet+5k')
def base_arch(pretrained=True):
return nn.Sequential(*list(self.model.children()))
self.base_arch = base_arch
def __init__(self,
dpn_version='dpn68',
input_size=128,
num_classes=340,
pretrained='imagenet',
dropout_rate=0.):
super(GeneralizedDPN, self).__init__()
if dpn_version == 'dpn68':
self.pretrained_model = dpn68(pretrained=pretrained)
self.fc_input = 832
elif dpn_version == 'dpn92':
self.pretrained_model = dpn92(
pretrained='imagenet+5k'
if pretrained else None) # TODO: ugly hard coded
self.fc_input = 2688
elif dpn_version == 'dpn98':
self.pretrained_model = dpn98(pretrained=pretrained)
self.fc_input = 2688
elif dpn_version == 'dpn107':
self.pretrained_model = dpn107(
pretrained='imagenet+5k'
if pretrained else None) # TODO: ugly hard coded
self.fc_input = 2688
elif dpn_version == 'dpn131':
self.pretrained_model = dpn131(pretrained=pretrained)
self.fc_input = 2688
else:
raise NotImplementedError('No implementation')
self.features = self.pretrained_model.features
self.relu = nn.ReLU()
self.avg_pool = nn.AvgPool2d(input_size // 32, stride=1, padding=0)
self.dropout = nn.Dropout(p=dropout_rate)
self.last_linear = nn.Linear(self.fc_input, num_classes)
def __init__(self, num_classes=2, arch="dpn92"):
super().__init__()
backbone = dpn92(pretrained='imagenet+5k').features
self.conv0 = nn.Sequential(backbone[0])
self.block1 = nn.Sequential(*backbone[1:4])
self.block2 = nn.Sequential(*backbone[4:8])
self.block3 = nn.Sequential(*backbone[8:28])
self.block4 = nn.Sequential(*backbone[28:])
def get_features(network):
if network == 'resnet50':
rn = resnet50(True)
features = nn.ModuleList([
nn.Sequential(rn.conv1, rn.bn1, rn.relu, rn.maxpool, rn.layer1),
rn.layer2, rn.layer3, rn.layer4
])
channels = 256, 512, 1024, 2048
elif network == 'resnext101_32x4d':
features = nn.ModuleList(list(resnext101_32x4d().features)[:8])
features = nn.ModuleList([
nn.Sequential(*features[:5]),
features[5],
features[6],
features[7],
])
channels = 256, 512, 1024, 2048
elif network == 'dpn68b':
dpn = list(dpn68b().features)
features = nn.ModuleList([
nn.Sequential(*dpn[:4]),
nn.Sequential(*dpn[4:8]),
nn.Sequential(*dpn[8:20]),
nn.Sequential(*dpn[20:23])
])
channels = 144, 320, 704, 832
elif network == 'dpn92':
dpn = list(dpn92().features)
features = nn.ModuleList([
nn.Sequential(*dpn[:4]),
nn.Sequential(*dpn[4:8]),
nn.Sequential(*dpn[8:28]),
nn.Sequential(*dpn[28:31])
])
channels = 336, 704, 1552, 2688
else:
raise ValueError(network)
return features, channels