YOLOv3学习笔记

代码为 https://github.com/Peterisfar/YOLOV3.git

神经网络输入尺寸 input_size
img.size()
torch.Size([1, 3, 544, 544])

从darknet53中获取特征

x_s, x_m, x_l = self.__backnone(x)
x_s.size()
torch.Size([1, 256, 68, 68])
x_m.size()
torch.Size([1, 512, 34, 34])
x_l.size()
torch.Size([1, 1024, 17, 17])

68=544/8 小尺寸

34=544/16 中尺寸

17=544/32 大尺寸

这里的8/16/32就是stride

MODEL = {"ANCHORS":[[(1.25, 1.625), (2.0, 3.75), (4.125, 2.875)],  # Anchors for small obj
            [(1.875, 3.8125), (3.875, 2.8125), (3.6875, 7.4375)],  # Anchors for medium obj
            [(3.625, 2.8125), (4.875, 6.1875), (11.65625, 10.1875)]] ,# Anchors for big obj
         "STRIDES":[8, 16, 32],
         "ANCHORS_PER_SCLAE":3
         }

x_s, x_m, x_l = self.__fpn(x_l, x_m, x_s)
x_s.size()
torch.Size([1, 75, 68, 68])
x_m.size()
torch.Size