amdegroot/ssd.pytorch 代码来源
一、vgg基础网络
网络的backbone是vgg，构建vgg网络代码如下，输入是vgg的各卷积层通道数和是否池化层的参数cfg，输入图像通道数i,
最后的conv6和conv7对应于上图中的Conv6和Conv7两个19*19*1024的特征图，其中Conv7用于预测
不得不说这个代码写的真棒
cfg = [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'C', 512, 512, 512, 'M',
512, 512, 512]
def vgg(cfg, i, batch_norm=False):
layers = []
in_channels = i
for v in cfg:
if v == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
elif v == 'C':
layers += [nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True)]
else:
conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
if batch_norm:
layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
else:
layers += [conv2d, nn.ReLU(inplace=True)]
in_channels = v
pool5 = nn.MaxPool2d(kernel_size=3, stride=1, padding=1)
conv6 = nn.Conv2d(512, 1024, kernel_size=3, padding=6, dilation=6)
conv7 = nn.Conv2d(1024, 1024, kernel_size=1)
layers += [pool5, conv6,
nn.ReLU(inplace=True), conv7, nn.ReLU(inplace=True)]
return layers二、额外的卷积层
SSD在基础的vgg网络后边添加了额外的卷积层，通过这些卷积层得到的feature map逐级减小，如图一中的19*19，10*10，5*5，3*3和1*1，
在这些层上边进行预测就可以得到多尺度的效果。构建代码如下，由于在之前的vgg函数中已经构建了Conv6和Conv7两层，
这里的输入就是Conv7得到的19*19*1024的feature map。cfg是要构建的卷积层的通道数，S表示这一层需要stride=2
cfg = [256, 'S', 512, 128, 'S', 256, 128, 256, 128, 256]
def add_extras(cfg, i, batch_norm=False):
# Extra layers added to VGG for feature scaling
# 1*1和3*3的卷积交替
layers = []
in_channels = i
flag = False
for k, v in enumerate(cfg):
if in_channels != 'S':
if v
== 'S':
layers += [nn.Conv2d(in_channels, cfg[k + 1],
kernel_size=(1, 3)[flag], stride=2, padding=1)]
else:
layers += [nn.Conv2d(in_channels, v, kernel_size=(1, 3)[flag])]
flag = not flag
in_channels = v
return layers三、multibox层
这一层是分类和位置回归层，对应于图一中的classifier。loc_layers由6个输出维度为default_box_num * 4的3*3卷积层组成，
conf_layers由6个输出维度为default_box_num * class_num的3*3卷积层组成
cfg = [4, 6, 6, 6, 4, 4] #每个预测层的default box个数
def multibox(vgg, extra_layers, cfg, num_classes):
loc_layers = [] #分别对loc回归层和conf预测层创建两个list
conf_layers = []
#vgg的layer中索引为21和-2的卷积输出的feature map对应是38*38*512和19*19*1024的预测层
vgg_source = [21, -2]
for k, v in enumerate(vgg_source):
loc_layers += [nn.Conv2d(vgg[v].out_channels,
cfg[k] * 4, kernel_size=3, padding=1)]
conf_layers += [nn.Conv2d(vgg[v].out_channels,
cfg[k] * num_classes, kernel_size=3, padding=1)]
for k, v in enumerate(extra_layers[1::2], 2):
#k从2开始
#在extra_layers的list中每个2个索引取一次，即取3*3conv层
loc_layers += [nn.Conv2d(v.out_channels, cfg[k]
* 4, kernel_size=3, padding=1)]
conf_layers += [nn.Conv2d(v.out_channels, cfg[k]
* num_classes, kernel_size=3, padding=1)]
return vgg, extra_layers, (loc_layers, conf_layers)四、default_box的生成
这一部分的代码在prior_box.py文件中的PriorBox类中。
对每一张特征图，按照不同的scale和ratio生成k个default boxes
def forward(self):
"""
测试中参数如下
feature_maps: <class 'list'>: [38, 19, 10, 5, 3, 1]
steps: <class 'list'>: [8, 16, 32, 64, 100, 300]
min_size: <class 'list'>: [30, 60, 111, 162, 213, 264]
max_size: <class 'list'>: [60, 111, 162, 213, 264, 315]
aspect_ratios: <class 'list'>: [[2], [2, 3], [2, 3], [2, 3], [2], [2]]
"""
mean = []
#遍历6个特征图
for k, f in enumerate(self.feature_maps):
#遍历特征图中每一个点
for i, j in product(range(f), repeat=2):
f_k = self.image_size / self.steps[k] #特征图一个点对应原图大小
# unit center x,y
#计算这一点上默认框的中心，如上边默认框中心设定的公式，将每一个点归一化
cx = (j + 0.5) / f_k
cy = (i + 0.5) / f_k
# aspect_ratio: 1，长宽比为1，用min_size作边长的默认框
# rel size: min_size
s_k = self.min_sizes[k]/self.image_size
mean += [cx, cy, s_k, s_k]
# aspect_ratio: 1，长宽比为1时额外添加一个尺度
# rel size: 计算公式sqrt(s_k * s_(k+1))，s_k+1等于这一层的max_size
s_k_prime = sqrt(s_k * (self.max_sizes[k]/self.image_size))
mean += [cx, cy, s_k_prime, s_k_prime]
# rest of aspect ratios，计算其他长宽比的默认框，w和h分别按照公式
for ar in self.aspect_ratios[k]:
mean += [cx, cy, s_k*sqrt(ar), s_k/sqrt(ar)]
mean += [cx, cy, s_k/sqrt(ar), s_k*sqrt(ar)]
# back to torch land，output是(8732,4)的Tensor
output = torch.Tensor(mean).view(-1, 4)
if self.clip:
output.clamp_(max=1, min=0) #控制output的范围在[0,1]
return output
