VGGM神经网络

目标跟踪论文一般提到VGG-M神经网络，也就是CNN-M神经网络，其出处是论文《Return of the Devil in the Details:
Delving Deep into Convolutional Nets》，其定义如下：

其架构包含5个卷积层和3个全连接层，它的特点是第一个卷积层的步幅减小和感受野较小，这在ILSVRC数据集上被证明是有益的。同时，conv2使用更大的步幅（stride=2而不是1）来保持合理的计算时间。还在conv4层使用更少的过滤器（512）。

有时候为了神经网络的鲁棒性和泛化性，也会在VGG-M中添加跨通道局部响应归一化（SpatialMapLRN）。具体代码如下：

class SpatialCrossMapLRN(nn.Module):
def __init__(self, local_size=1, alpha=1.0, beta=0.75, k=1, ACROSS_CHANNELS=True):
super(SpatialCrossMapLRN, self).__init__()
self.ACROSS_CHANNELS = ACROSS_CHANNELS
if ACROSS_CHANNELS:
self.average=nn.AvgPool3d(kernel_size=(local_size, 1, 1),
stride=1,
padding=(int((local_size-1.0)/2), 0, 0))
else:
self.average=nn.AvgPool2d(kernel_size=local_size,
stride=1,
padding=int((local_size-1.0)/2))
self.alpha = alpha
self.beta = beta
self.k = k
def forward(self, x):
if self.ACROSS_CHANNELS:
div = x.pow(2).unsqueeze(1)
div = self.average(div).squeeze(1)
div = div.mul(self.alpha).add(self.k).pow(self.beta)
else:
div = x.pow(2)
div = self.average(div)
div = div.mul(self.alpha).add(self.k).pow(self.beta)
x = x.div(div)
return x
class VGGM(nn.Module):
def __init__(self, num_classes=1000):
super(VGGM, self).__init__()
self.num_classes = num_classes
self.features = nn.Sequential(
nn.Conv2d(3, 96, (7, 7), (2, 2)), # conv1
nn.ReLU(),
SpatialCrossMapLRN(5, 0.0005, 0.75, 2),
nn.MaxPool2d((3, 3), (2, 2), (0, 0), ceil_mode=True),
nn.Conv2d(96, 256, (5, 5), (2, 2), (1, 1)), # conv2
nn.ReLU(),
SpatialCrossMapLRN(5, 0.0005, 0.75, 2),
nn.MaxPool2d((3, 3), (2, 2), (0, 0), ceil_mode=True),
nn.Conv2d(256, 512, (3, 3), (1, 1), (1, 1)), # conv3
nn.ReLU(),
nn.Conv2d(512, 512, (3, 3), (1, 1), (1, 1)), # conv4
nn.ReLU(),
nn.Conv2d(512, 512, (3, 3), (1, 1), (1, 1)), # conv5
nn.ReLU(),
nn.MaxPool2d((3, 3), (2, 2), (0, 0), ceil_mode=True)
)
self.classifier = nn.Sequential(
nn.Linear(18432, 4096),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(4096, 4096),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(4096, num_classes)
)
def forward(self, x):
x = self.features(x)
x = x.view(x.size(0), -1)
x = self.classifier(x)
return x
def vggm(num_classes=1000, pretrained='imagenet'):
if pretrained:
settings = pretrained_settings['vggm'][pretrained]
assert num_classes == settings['num_classes'], \
"num_classes should be {}, but is {}".format(settings['num_classes'], num_classes)
model = VGGM(num_classes=num_classes)
model.load_state_dict(torch.load('../vggm.pth', map_location=lambda storage, loc: storage))
model.input_space = settings['input_space']
model.input_size = settings['input_size']
model.input_range = settings['input_range']
model.mean = settings['mean']
model.std = settings['std']
else:
model = VGGM(num_classes=num_classes)
return model