def __init__(self, config):
super(YoloBody, self).__init__()
self.config = config
# backbone，从cfg改为直接init
self.backbone = darknet53(None)
out_filters = self.backbone.layers_out_filters
# last_layer0
final_out_filter0 = len(
config["yolo"]["anchors"][0]) * (5 + config["yolo"]["classes"])
self.last_layer0 = make_last_layers([512, 1024], out_filters[-1],
final_out_filter0)
# embedding1
final_out_filter1 = len(
config["yolo"]["anchors"][1]) * (5 + config["yolo"]["classes"])
self.last_layer1_conv = conv2d(512, 256, 1)
self.last_layer1_upsample = nn.Upsample(scale_factor=2,
mode='nearest') #26, 26, 256
#堆叠前两个
self.last_layer1 = make_last_layers([256, 512], out_filters[-2] + 256,
final_out_filter1)
# embedding2
final_out_filter2 = len(
config["yolo"]["anchors"][2]) * (5 + config["yolo"]["classes"])
self.last_layer2_conv = conv2d(256, 128, 1)
self.last_layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.last_layer2 = make_last_layers([128, 256], out_filters[-3] + 128,
final_out_filter2)
def __init__(self, config):
super(YoloBody, self).__init__()
self.config = config
# backbone
self.backbone = darknet53(None) # 获取darknet结构，保存到 self.backbone
out_filters = self.backbone.layers_out_filters # layers_out_filters = [64, 128, 256, 512, 1024]
# last_layer0 # 3*(5+num_classes) = 3*(5+20)=3*(4+1+20)=75
final_out_filter0 = len(config["yolo"]["anchors"][0]) * (
5 + config["yolo"]["classes"]
) # 3*(5+num_classes) = 3*(5+20)=3*(4+1+20)=75
self.last_layer0 = make_last_layers(
[512, 1024], out_filters[-1], final_out_filter0) # 7次卷积（5次卷积+2次卷积）
# embedding1 75
final_out_filter1 = len(config["yolo"]["anchors"][1]) * (
5 + config["yolo"]["classes"]) # 75
self.last_layer1_conv = conv2d(512, 256, 1) # 卷积
self.last_layer1_upsample = nn.Upsample(
scale_factor=2, mode='nearest') # 上采样，高宽扩张为26x26
# 26，26，256
self.last_layer1 = make_last_layers(
[256, 512], out_filters[-2] + 256,
final_out_filter1) # 7次卷积（5次卷积+2次卷积）
# embedding2 75
final_out_filter2 = len(config["yolo"]["anchors"][2]) * (
5 + config["yolo"]["classes"]) # 75
self.last_layer2_conv = conv2d(256, 128, 1)
self.last_layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
# 52，52，128
self.last_layer2 = make_last_layers(
[128, 256], out_filters[-3] + 128,
final_out_filter2) # 堆叠完成后， 7次卷积（5次卷积+2次卷积）
def __init__(self, config):
super(YoloBody, self).__init__()
self.config = config
#---------------------------------------------------#
# 生成darknet53的主干模型
# 获得三个有效特征层，他们的shape分别是：
# 52,52,256
# 26,26,512
# 13,13,1024
#---------------------------------------------------#
self.backbone = darknet53(None)
# out_filters : [64, 128, 256, 512, 1024]
out_filters = self.backbone.layers_out_filters
#------------------------------------------------------------------------#
# 计算yolo_head的输出通道数，对于voc数据集而言
# final_out_filter0 = final_out_filter1 = final_out_filter2 = 75
#------------------------------------------------------------------------#
final_out_filter0 = len(config["yolo"]["anchors"][0]) * (5 + config["yolo"]["classes"])
self.last_layer0 = make_last_layers([512, 1024], out_filters[-1], final_out_filter0)
final_out_filter1 = len(config["yolo"]["anchors"][1]) * (5 + config["yolo"]["classes"])
self.last_layer1_conv = conv2d(512, 256, 1)
self.last_layer1_upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.last_layer1 = make_last_layers([256, 512], out_filters[-2] + 256, final_out_filter1)
final_out_filter2 = len(config["yolo"]["anchors"][2]) * (5 + config["yolo"]["classes"])
self.last_layer2_conv = conv2d(256, 128, 1)
self.last_layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.last_layer2 = make_last_layers([128, 256], out_filters[-3] + 128, final_out_filter2)
def __init__(self, config):
super(YoloBody, self).__init__()
self.config = config
# backbone
self.backbone = darknet53(None) #获取darknext53 的结构，保存在self.backbone
out_filters = self.backbone.layers_out_filters
# last_layer0 3*(5+num_classes)=3*(5+20)=3*(4+1+20)=75
final_out_filter0 = len(
config["yolo"]["anchors"][0]) * (5 + config["yolo"]["classes"])
self.last_layer0 = make_last_layers([512, 1024], out_filters[-1],
final_out_filter0)
# embedding1 75
final_out_filter1 = len(
config["yolo"]["anchors"][1]) * (5 + config["yolo"]["classes"])
self.last_layer1_conv = conv2d(512, 256, 1)
self.last_layer1_upsample = nn.Upsample(scale_factor=2, mode='nearest')
# 获得一个26*26*256的特征层
self.last_layer1 = make_last_layers([256, 512], out_filters[-2] + 256,
final_out_filter1)
# embedding2 75
final_out_filter2 = len(
config["yolo"]["anchors"][2]) * (5 + config["yolo"]["classes"])
self.last_layer2_conv = conv2d(256, 128, 1)
self.last_layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
# 52*52*128
self.last_layer2 = make_last_layers([128, 256], out_filters[-3] + 128,
final_out_filter2)
def __init__(self, anchor, num_classes):
"""
YOLOv3 网络结构初始化\n
"""
super(YoloBody, self).__init__()
#---------------------------------------------------#
# 生成 darknet53的主干模型。
# 从 darknet53网络输出中，获得三个有效特征层。
# 他们的 shape分别是：
# 52,52,256
# 26,26,512
# 13,13,1024
#---------------------------------------------------#
self.backbone = darknet53(None) # 默认不装载任何预训练模型
# backbone_out_channels : [64, 128, 256, 512, 1024]
backbone_out_channels = self.backbone.layers_out_channels
# 特征金字塔采样并输出
# 网络输出：13* 13* final_out_channel0
final_out_channel0 = len(anchor[0]) * (
5 + num_classes) # 单个 grid cell 最多可侦测 len(anchor[0])个物体
self.final_layer0 = make_last_layers(
[512, 1024], backbone_out_channels[-1], final_out_channel0
) # channels: 1024 --> final_out_channel0; 不需要 concat
# 网络输出：26* 26* final_out_channel1
self.layer1_conv = conv2d(512, 256,
1) # 1* 1 Conv, 13* 13* 512 --> 13* 13* 256
self.layer1_upsample = nn.Upsample(
scale_factor=2, mode='nearest') # 上采样：13* 13* 256 --> 26* 26* 256
final_out_channel1 = len(anchor[1]) * (5 + num_classes)
self.final_layer1 = make_last_layers(
[256, 512], backbone_out_channels[-2] + 256, final_out_channel1
) # channels: 768 --> final_out_channel1; 768 是经过 concat后的尺寸
# 网络输出：52* 52* final_out_channel2
self.layer2_conv = conv2d(256, 128, 1)
self.layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
final_out_channel2 = len(anchor[2]) * (5 + num_classes)
self.final_layer2 = make_last_layers(
[128, 256], backbone_out_channels[-3] + 128, final_out_channel2
) # channels: 384 --> final_out_channel2; 384 是经过 concat后的尺寸
def __init__(self, anchors_mask, num_classes, pretrained=False):
super(YoloBody, self).__init__()
#---------------------------------------------------#
# 生成darknet53的主干模型
# 获得三个有效特征层，他们的shape分别是：
# 52,52,256
# 26,26,512
# 13,13,1024
#---------------------------------------------------#
self.backbone = darknet53()
if pretrained:
self.backbone.load_state_dict(
torch.load("model_data/darknet53_backbone_weights.pth"))
#---------------------------------------------------#
# out_filters : [64, 128, 256, 512, 1024]
#---------------------------------------------------#
out_filters = self.backbone.layers_out_filters
#------------------------------------------------------------------------#
# 计算yolo_head的输出通道数，对于voc数据集而言
# final_out_filter0 = final_out_filter1 = final_out_filter2 = 75
#------------------------------------------------------------------------#
self.last_layer0 = make_last_layers([512, 1024], out_filters[-1],
len(anchors_mask[0]) *
(num_classes + 5))
self.last_layer1_conv = conv2d(512, 256, 1)
self.last_layer1_upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.last_layer1 = make_last_layers([256, 512], out_filters[-2] + 256,
len(anchors_mask[1]) *
(num_classes + 5))
self.last_layer2_conv = conv2d(256, 128, 1)
self.last_layer2_upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.last_layer2 = make_last_layers([128, 256], out_filters[-3] + 128,
len(anchors_mask[2]) *
(num_classes + 5))
def __init__(self, config):
super(YoloBody, self).__init__()
self.config = config
# backbone
self.backbone = darknet53(
None) # 将darknet.py中获得的主干网络的结构保存在.backbone属性中。
out_filters = self.backbone.layers_out_filters
# last_layer0 3*（5+num_classes)=3*(5+20)=3*(4+1+20)=75 这部分是处理out5的特征层
final_out_filter0 = len(config["yolo"]["anchors"][0]) * (
5 + config["yolo"]["classes"]) # final_out_filter0就是75 是特征图的参数
self.last_layer0 = make_last_layers(
[512, 1024], out_filters[-1],
final_out_filter0) # make_last_layers是七次卷积，最后两次卷积是回归预测和分类预测
# embedding1 75 这部分是处理out4的特征层
final_out_filter1 = len(
config["yolo"]["anchors"][1]) * (5 + config["yolo"]["classes"])
self.last_layer1_conv = conv2d(512, 256, 1) # 用1*1的卷积调整通道数
self.last_layer1_upsample = nn.Upsample(scale_factor=2,
mode='nearest') # 第一次上采样
# 此处已经获得26,26,256的特征层
self.last_layer1 = make_last_layers(
[256, 512], out_filters[-2] + 256, final_out_filter1
) # 在前向传播时两个尺度的特征层进行了堆叠。make_last_layers是七次卷积，最后两次卷积是回归预测和分类预测
# embedding2 75 这部分是处理out3的特征层
final_out_filter2 = len(
config["yolo"]["anchors"][2]) * (5 + config["yolo"]["classes"])
self.last_layer2_conv = conv2d(256, 128, 1) # 1*1卷积调整通道数
self.last_layer2_upsample = nn.Upsample(scale_factor=2,
mode='nearest') # 第二次上采样
# 此处已经获得52,52,128的特征层
self.last_layer2 = make_last_layers(
[128, 256], out_filters[-3] + 128, final_out_filter2
) # 在前向传播时两个尺度的特征层进行了堆叠。make_last_layers是七次卷积，最后两次卷积是回归预测和分类预测
