def Split_ResNet101(cfg, progress=True):
model = ResNet(cfg, get_builder(cfg), Bottleneck, [3, 4, 23, 3])
if cfg.pretrained == 'imagenet':
arch = 'resnet101'
state_dict = load_state_dict_from_url(model_urls[arch],
progress=progress)
load_state_dict(model, state_dict, strict=False)
return model
def Split_ResNet18(cfg, progress=True):
model = ResNet(cfg, get_builder(cfg), BasicBlock, [2, 2, 2, 2])
if cfg.pretrained == 'imagenet':
arch = 'resnet18'
state_dict = load_state_dict_from_url(model_urls[arch],
progress=progress)
load_state_dict(model, state_dict, strict=False)
return model
def Split_densenet201(cfg, pretrained=False, progress=True, **kwargs):
r"""Densenet-201 model from
`"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
progress (bool): If True, displays a progress bar of the download to stderr
memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
"""
return _densenet(cfg, get_builder(cfg), 'densenet201', 32, (6, 12, 48, 32),
64, pretrained, progress, **kwargs)
def __init__(self,cfg, num_classes=1000, aux_logits=False, transform_input=False, init_weights=None,
blocks=None): # AT : I disabled the aux_logits
super(GoogLeNet, self).__init__()
if blocks is None:
blocks = [BasicConv2d, Inception, InceptionAux]
if init_weights is None:
warnings.warn('The default weight initialization of GoogleNet will be changed in future releases of '
'torchvision. If you wish to keep the old behavior (which leads to long initialization times'
' due to scipy/scipy#11299), please set init_weights=True.', FutureWarning)
init_weights = True
assert len(blocks) == 3
builder = get_builder(cfg)
slim_factor = cfg.slim_factor
if slim_factor < 1:
cfg.logger.info('WARNING: You are using a slim network')
conv_block = blocks[0]
inception_block = blocks[1]
inception_aux_block = blocks[2]
self.aux_logits = aux_logits
self.transform_input = transform_input
slim = lambda x: math.ceil(x * slim_factor)
self.conv1 = conv_block(builder,3, slim(64), kernel_size=7, stride=2) # , padding=3
self.maxpool1 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.conv2 = conv_block(builder,slim(64), slim(64), kernel_size=1)
self.conv3 = conv_block(builder,slim(64), slim(192), kernel_size=3) # padding=1
self.maxpool2 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception3a = inception_block(builder, slim(192), slim(64)
, slim(96), slim(128), slim(16), slim(32), slim(32))
prev_out_channels = self.inception3a.out_channels # This is 256
concat_order = self.inception3a.concat_order
self.inception3b = inception_block(builder,prev_out_channels, slim(128), slim(128), slim(192), slim(32), slim(96), slim(64),in_channels_order=concat_order)
prev_out_channels = self.inception3b.out_channels # This is 480
concat_order = self.inception3b.concat_order
self.maxpool3 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception4a = inception_block(builder,prev_out_channels, slim(192), slim(96), slim(208), slim(16), slim(48), slim(64),in_channels_order=concat_order)
prev_out_channels = self.inception4a.out_channels # This is 512
concat_order = self.inception4a.concat_order
self.inception4b = inception_block(builder,prev_out_channels, slim(160), slim(112), slim(224), slim(24), slim(64), slim(64),in_channels_order=concat_order)
prev_out_channels = self.inception4b.out_channels # This is 512
concat_order = self.inception4b.concat_order
self.inception4c = inception_block(builder,prev_out_channels, slim(128), slim(128), slim(256), slim(24), slim(64), slim(64),in_channels_order=concat_order)
prev_out_channels = self.inception4c.out_channels # This is 512
concat_order = self.inception4c.concat_order
self.inception4d = inception_block(builder,prev_out_channels, slim(112), slim(144), slim(288), slim(32), slim(64), slim(64),in_channels_order=concat_order)
prev_out_channels = self.inception4d.out_channels # This is 528
concat_order = self.inception4d.concat_order
self.inception4e = inception_block(builder,prev_out_channels, slim(256), slim(160), slim(320), slim(32), slim(128), slim(128),in_channels_order=concat_order)
prev_out_channels = self.inception4e.out_channels # This is 832
concat_order = self.inception4e.concat_order
self.maxpool4 = nn.MaxPool2d(2, stride=2, ceil_mode=True)
self.inception5a = inception_block(builder,prev_out_channels, slim(256), slim(160), slim(320), slim(32), slim(128), slim(128),in_channels_order=concat_order)
prev_out_channels = self.inception5a.out_channels # This is 832
concat_order = self.inception5a.concat_order
self.inception5b = inception_block(builder,prev_out_channels, slim(384), slim(192), slim(384), slim(48), slim(128), slim(128),in_channels_order=concat_order)
prev_out_channels = self.inception5b.out_channels # This is 1024
concat_order = self.inception5b.concat_order
if aux_logits:
self.aux1 = inception_aux_block(builder,512, num_classes)
self.aux2 = inception_aux_block(builder,528, num_classes)
else:
self.aux1 = None
self.aux2 = None
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.dropout = nn.Dropout(0.2)
# self.fc = nn.Linear(1024, num_classes)
self.fc = builder.linear(prev_out_channels, num_classes,last_layer=True,in_channels_order=concat_order)
if init_weights:
self._initialize_weights()