def __init__(self, jigsaw_classes=1000, n_classes=100, domains=3, dropout=True):
super(AlexNetCaffe, self).__init__()
print("Using Caffe AlexNet")
self.features_1 = nn.Sequential(OrderedDict([
("conv1", nn.Conv2d(3, 96, kernel_size=11, stride=4)),
("relu1", nn.ReLU(inplace=True)),]))
self.infodrop_1 = Info_Dropout(3, 96, kernel_size=11, stride=4)
self.features = nn.Sequential(OrderedDict([
("pool1", nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)),
("norm1", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv2", nn.Conv2d(96, 256, kernel_size=5, padding=2, groups=2)),
("relu2", nn.ReLU(inplace=True)),
("pool2", nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)),
("norm2", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv3", nn.Conv2d(256, 384, kernel_size=3, padding=1)),
("relu3", nn.ReLU(inplace=True)),
("conv4", nn.Conv2d(384, 384, kernel_size=3, padding=1, groups=2)),
("relu4", nn.ReLU(inplace=True)),
("conv5", nn.Conv2d(384, 256, kernel_size=3, padding=1, groups=2)),
("relu5", nn.ReLU(inplace=True)),
("pool5", nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)),
]))
self.classifier = nn.Sequential(OrderedDict([
("fc6", nn.Linear(256 * 6 * 6, 4096)),
("relu6", nn.ReLU(inplace=True)),
("drop6", nn.Dropout() if dropout else Id()),
("fc7", nn.Linear(4096, 4096)),
("relu7", nn.ReLU(inplace=True)),
("drop7", nn.Dropout() if dropout else Id())]))
self.jigsaw_classifier = nn.Linear(4096, jigsaw_classes)
self.class_classifier = nn.Linear(4096, n_classes)
def __init__(self, jigsaw_classes=1000, n_classes=100, dropout=True):
super().__init__()
print("Using Caffe AlexNet")
self.features = nn.Sequential(
OrderedDict([
("conv1", nn.Conv2d(3, 96, kernel_size=11, stride=2)),
("relu1", nn.ReLU(inplace=True)),
("pool1", nn.MaxPool2d(kernel_size=3, stride=2,
ceil_mode=True)),
("norm1", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv2", nn.Conv2d(96,
256,
kernel_size=5,
padding=2,
groups=2)),
("relu2", nn.ReLU(inplace=True)),
("pool2", nn.MaxPool2d(kernel_size=3, stride=2,
ceil_mode=True)),
("norm2", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv3", nn.Conv2d(256, 384, kernel_size=3, padding=1)),
("relu3", nn.ReLU(inplace=True)),
("conv4",
nn.Conv2d(384, 384, kernel_size=3, padding=1, groups=2)),
("relu4", nn.ReLU(inplace=True)),
("conv5",
nn.Conv2d(384, 256, kernel_size=3, padding=1, groups=2)),
("relu5", nn.ReLU(inplace=True)),
# ("pool5", nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)),
]))
self.fc_size = 4096
self.classifier = nn.Sequential(
OrderedDict([("fc6", nn.Linear(256 * 6 * 6, self.fc_size)),
("relu6", nn.ReLU(inplace=True)),
("drop6", nn.Dropout() if dropout else Id()),
("fc7", nn.Linear(4096, 4096)),
("relu7", nn.ReLU(inplace=True)),
("drop7", nn.Dropout() if dropout else Id())]))
self.jigsaw_classifier = nn.Sequential(
nn.Linear(9 * self.fc_size, jigsaw_classes),
# nn.ReLU(inplace=True),
# nn.Dropout(),
# nn.Linear(4096, jigsaw_classes)
)
self.class_classifier = nn.Sequential(
nn.Linear(4096, n_classes),
# nn.ReLU(inplace=True),
# nn.Dropout(),
# nn.Linear(4096, n_classes)
)
for m in self.modules():
if isinstance(m, nn.Linear):
nn.init.xavier_uniform_(m.weight, .1)
nn.init.constant_(m.bias, 0.)
def __init__(self, n_classes=100, domains=3, dropout=True):
super(AlexNetCaffe, self).__init__()
print("Using Caffe AlexNet")
self.features = nn.Sequential(
OrderedDict([
("conv1", nn.Conv2d(3, 96, kernel_size=11, stride=4)),
("relu1", nn.ReLU(inplace=True)),
("pool1", nn.MaxPool2d(kernel_size=3, stride=2,
ceil_mode=True)),
("norm1", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv2", nn.Conv2d(96,
256,
kernel_size=5,
padding=2,
groups=2)),
("relu2", nn.ReLU(inplace=True)),
("pool2", nn.MaxPool2d(kernel_size=3, stride=2,
ceil_mode=True)),
("norm2", nn.LocalResponseNorm(5, 1.e-4, 0.75)),
("conv3", nn.Conv2d(256, 384, kernel_size=3, padding=1)),
("relu3", nn.ReLU(inplace=True)),
("conv4",
nn.Conv2d(384, 384, kernel_size=3, padding=1, groups=2)),
("relu4", nn.ReLU(inplace=True)),
("conv5",
nn.Conv2d(384, 256, kernel_size=3, padding=1, groups=2)),
("relu5", nn.ReLU(inplace=True)),
("pool5", nn.MaxPool2d(kernel_size=3, stride=2,
ceil_mode=True)),
]))
self.classifier = nn.Sequential(
OrderedDict([("fc6", nn.Linear(256 * 6 * 6, 4096)),
("relu6", nn.ReLU(inplace=True)),
("drop6", nn.Dropout() if dropout else Id()),
("fc7", nn.Linear(4096, 4096)),
("relu7", nn.ReLU(inplace=True)),
("drop7", nn.Dropout() if dropout else Id())]))
classes = n_classes
K = 2
self.sms = torch.nn.Parameter(torch.normal(0.,
1e-3,
size=[K, 4096, classes],
dtype=torch.float,
device='cuda'),
requires_grad=True)
self.sm_biases = torch.nn.Parameter(torch.normal(0.,
1e-3,
size=[K, classes],
dtype=torch.float,
device='cuda'),
requires_grad=True)
self.embs = torch.nn.Parameter(torch.normal(mean=0.,
std=1e-1,
size=[3, K - 1],
dtype=torch.float,
device='cuda'),
requires_grad=True)
self.cs_wt = torch.nn.Parameter(torch.normal(mean=0.,
std=1e-4,
size=[],
dtype=torch.float,
device='cuda'),
requires_grad=True)