yolov5s网络结构

Yolov5s是基于PyTorch实现的目标检测模型，它的网络结构如下：

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Focus-1             [-1, 80, 80, 80]               0
     ConvBlock-2            [-1, 160, 40, 40]          11,840
            C3-3            [-1, 320, 40, 40]         155,680
     ConvBlock-4            [-1, 640, 20, 20]         161,280
            C3-5           [-1, 1280, 20, 20]         621,440
     ConvBlock-6            [-1, 640, 20, 20]         819,840
            C3-7           [-1, 1280, 20, 20]       2,483,200
     ConvBlock-8            [-1, 640, 20, 20]       1,638,400
            C3-9           [-1, 1280, 20, 20]       4,966,400
           SPP-10           [-1, 1280, 20, 20]               0
     ConvBlock-11            [-1, 640, 20, 20]       8,194,560
           C3-12           [-1, 1280, 20, 20]      19,865,600
DetectionBlock-13  [[-1, 80, 80, 20], [-1, 40, 40, 20], [-1, 20, 20, 20]]               0
           C3-14           [-1, 512, 40, 40]       6,386,688
     ConvBlock-15            [-1, 256, 40, 40]       1,318,144
         Upsample-16            [-1, 256, 80, 80]               0
           C3-17            [-1, 256, 80, 80]       1,596,544
DetectionBlock-18  [[-1, 80, 80, 20], [-1, 40, 40, 20], [-1, 20, 20, 20]]               0
================================================================
Total params: 47,107,776
Trainable params: 47,107,776
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.88
Forward/backward pass size (MB): 466.06
Params size (MB): 179.54
Estimated Total Size (MB): 647.49
----------------------------------------------------------------

其中，`Focus`模块是YOLOv5中新增的模块，它将输入的特征图进行通道和空间的压缩，以减少模型的计算量。`C3`模块是YOLOv5中的核心模块，它由三层卷积组成，并且引入了残差结构和SPP结构。`DetectionBlock`模块是YOLOv5中的检测模块，它用于生成检测框和预测类别。`Upsample`模块是YOLOv5中新增的模块，用于将特征图进行上采样。