论文地址：https://arxiv.org/pdf/1708.02002.pdf
pytorch实现:https://github.com/yhenon/pytorch-retinanet

网络模型

RetinaNet主要由ResNet网络，FPN网络和两个子网络构成。其中ResNet下采样，提取不同level的浅层特征。FPN又将feature map进行上采样得到深层特征。将深层特征与浅层特征相加得到不同的feature map，即P3,P4,P5。然后两个子网络在提取的feature map进行预测，class子网络预测类别，box子网络预测边界框。两个子网络虽然结构相似，但是不共享网络权重。各个功能模块的具体细节将在下面结合代码进行说明。

RetinaNet网络模型

ResNet50网络模型

ResNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU(inplace=True)
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (layer1): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer2): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer3): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (4): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (5): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer4): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (fpn): PyramidFeatures(
    (P5_1): Conv2d(2048, 256, kernel_size=(1, 1), stride=(1, 1))
    (P5_upsampled): Upsample(scale_factor=2.0, mode=nearest)
    (P5_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P4_1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
    (P4_upsampled): Upsample(scale_factor=2.0, mode=nearest)
    (P4_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P3_1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
    (P3_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P6): Conv2d(2048, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    (P7_1): ReLU()
    (P7_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
  )
  (regressionModel): RegressionModel(
    (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act1): ReLU()
    (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act2): ReLU()
    (conv3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act3): ReLU()
    (conv4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act4): ReLU()
    (output): Conv2d(256, 36, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  )
  (classificationModel): ClassificationModel(
    (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act1): ReLU()
    (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act2): ReLU()
    (conv3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act3): ReLU()
    (conv4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act4): ReLU()
    (output): Conv2d(256, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (output_act): Sigmoid()
  )
  (anchors): Anchors()
  (regressBoxes): BBoxTransform()
  (clipBoxes): ClipBoxes()
  (focalLoss): FocalLoss()
)

以一张608x608的彩色图片作为输入，tensor为[batch_size,C,H,W]，故维度为[1,3,608,608]。
一、下采样和最大池化。
维度变化为[1,3,608,608]--conv1-->[1,64,304,304]--maxpool-->[1,64,152,152]

(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)

二、layer1,layer2,layer3,layer4卷积获得不同level的feature map。
其中layer2的fature map对应于上图的C3，layer3对应C4，layer 4对应C5。以获得C3为例，layer1得到的输出为[1,256,152,152]，layer2的前向传播流程图如下，得到的feature map C3维度为[1,512,76,76]。

C3的前向传播过程
下采样得到各层的feature map维度为：C3为[1,512,76,76]，C4为[1,1024,38,38]，C5为[1,2048,19,19]。
三、FPN网络将深层特征与浅层特征进行结合。对关于FPN的作用可以参考该篇博客(https://blog.csdn.net/WZZ18191171661/article/details/79494534
)。
在第一张图，网络模型中可以看出P5是由C5得到。而P5上采样之后的结果和C4进行1x1卷积之后的结果由于维度相同，故可以直接相加得到P4。同理，P3是由P4和C3组合得到。

(fpn): PyramidFeatures(
    (P5_1): Conv2d(2048, 256, kernel_size=(1, 1), stride=(1, 1))
    (P5_upsampled): Upsample(scale_factor=2.0, mode=nearest)
    (P5_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P4_1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
    (P4_upsampled): Upsample(scale_factor=2.0, mode=nearest)
    (P4_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P3_1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
    (P3_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (P6): Conv2d(2048, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    (P7_1): ReLU()
    (P7_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
  )

P3,P4,P5流程图

在源码中可以看到还有P6,P7的feature map，但是在论文图中并没有画出，P6,P7分别由P5,P7下采样得到，P6的维度为[1,256,10,10]，P7的维度为[1,256,5,5]。
四、在P3,P4...P7上进行类别预测和边界框预测
由于类别预测与边界框预测的模型结构类似，故只需讲解一个即可。以在P3上进行边界框预测为例。维度变化为
P3[1,256,76,76]--conv1-->[1,256,76,76]--conv2-->[1,256,76,76]--conv3-->[1,256,76,76]--conv4-->[1,256,76,76]--output-->[1,36,76,76]--permute-->[1,76,76,36]--view-->[1,51984,4]
P3 feature map的大小为76x76，每个单元格预测9种不同比例大小的anchor box，故在该feature map上一共有51984（76x76x9）个anchor box。又由于有P3,P4...P7一共5个level的feature map，故一幅图片一共产生(76x76+38x38+19x19+10x10+5x5)x9=69354个anchor box。regressionModel后输出的维度为[1,69354,4]。4维向量为下图种的[tx,ty,tw,th]。同理，因为COCO的类别有80种，classificationModel后输出的维度为[1,69354,80]。

(regressionModel): RegressionModel(
    (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act1): ReLU()
    (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act2): ReLU()
    (conv3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act3): ReLU()
    (conv4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (act4): ReLU()
    (output): Conv2d(256, 36, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  )

转化公式
五、生成不同featrue map上的anchor box
不同level的feature map对应的原图大小不同。例如P3[1,256,76,76]的一个单元格对应原图的大小为608/76=8，而P4[1,256,38,38]对应大小为16，显然P3比P4更适合检测小目标。

def generate_anchors(base_size=16, ratios=None, scales=None):
    """
    Generate anchor (reference) windows by enumerating aspect ratios X
    scales w.r.t. a reference window.
    """
    if ratios is None:
        ratios = np.array([0.5, 1, 2])
    if scales is None:
        scales = np.array([2 ** 0, 2 ** (1.0 / 3.0), 2 ** (2.0 / 3.0)])
    num_anchors = len(ratios) * len(scales)  #一个单元格中的anchor box个数
    # initialize output anchors
    anchors = np.zeros((num_anchors, 4))
    # scale base_size
    anchors[:, 2:] = base_size * np.tile(scales, (2, len(ratios))).T
    # compute areas of anchors
    areas = anchors[:, 2] * anchors[:, 3]
    # correct for ratios
    anchors[:, 2] = np.sqrt(areas / np.repeat(ratios, len(scales)))
    anchors[:, 3] = anchors[:, 2] * np.repeat(ratios, len(scales))
    # transform from (x_ctr, y_ctr, w, h) -> (x1, y1, x2, y2)
    anchors[:, 0::2] -= np.tile(anchors[:, 2] * 0.5, (2, 1)).T
    anchors[:, 1::2] -= np.tile(anchors[:, 3] * 0.5, (2, 1)).T
    return anchors

不同大小和比例的anchor box

六、计算损失

该篇论文的主要创新是其损失函数，论文中提到网络之所以达到好的结果，不是因为网络结构设计的创新，而是损失函数的创新。我们都知道one-stage的检测速度快但是精度低。而two-stage的精度高但是检测速度慢。
上述原因:这是因为我们的网络在预测时，会产生了大量的anchor box，而这些anchor box绝大多是背景，这就导致了正负样本的不平衡。two-stage在第一阶段会对anchor box进行二分类，判别是背景还是前景，这样就大大减少了负样本的数量，然后在第二阶段进行类别的细分，判别前景到底属于哪个类别。
解决方法:为了解决由于正负样本不平衡而导致one-stage检测方法精度底下的问题，本文引入了新的损失函数(Focal loss)，是对标准的交叉熵损失函数的改进。其格式为下图，论文中得到最佳的a=0.25，γ =2。

Focal Loss计算公式
单个样本类别损失函数
对于边界框损失，只需考虑正样本的情况。其计算已经在上图的转化公式中给出。将类别损失与边界框损失相加得到总损失。

总结

RetinaNet一针见血的指出了one-stage检测与two-stage检测在精确度方面的差异是由正负样本不平衡所引起的。为了能够解决该问题，提高one-stage的检测性能。创新性的提出了Focal Loss，能够将往往易区分样本的loss大幅下降，而难以区分样本的loss几乎不变，从而使得模型更加注重难以区分的样本，提高检测性能。

参考博客：

FPN网络：https://blog.csdn.net/WZZ18191171661/article/details/79494534
RetinaNet:https://blog.csdn.net/JNingWei/article/details/80038594