yolov7代码梳理，读取json格式数据
Posted 东东就是我
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了yolov7代码梳理，读取json格式数据相关的知识，希望对你有一定的参考价值。
1.整体结构
说实话，看配置文件的方式真的不适合阅读代码，自己改写了一下
https://blog.csdn.net/u012863603/article/details/126118799 ，对着这个网络图
# coding=utf-8
import math
import torch
import torch.nn as nn
'''
Author:Don
date:2022/10/17 15:51
desc:
'''
from baseblock import *
from utils import *
class Model(nn.Module):
def __init__(self,classes=2,anchors=None,ch=3):
super(Model, self).__init__()
# self.train=False
na=3*(classes+5)
self.layer0=Conv(ch,32,3,1) # 0
self.layer1=Conv(32,64,3,2) # 1-p1/2
self.layer2=Conv(64,64,3,1)
self.layer3=Conv(64,128,3,2) #3-p2/4
self.layer11=ELAN(128)
self.layer16=MP_1(256)
self.layer24=ELAN(256)
self.layer29=MP_1(512)
self.layer37=ELAN(512)
self.layer42=MP_1(1024)
self.layer50=ELAN_E(1024)
self.layer51=SPPCSPC(1024,512)
self.layer52=Conv(512,256,1,1)
self.upsample=nn.Upsample(None,2,'nearest')
self.concat = Concat(1)
self.layer54=Conv(1024,256,1,1)
self.layer63=ELAN_W(512)
self.layer64=Conv(256,128,1,1)
self.layer66=Conv(512,128,1,1)
self.layer75=ELAN_W(256)
self.layer80=MP_2(128)
self.layer88=ELAN_W(512)
self.layer93=MP_2(256)
self.layer101=ELAN_W(1024)
self.layer102=RepConv(128,256,3,1)
self.layer103=RepConv(256,512,3,1)
self.layer104=RepConv(512,1024,3,1)
self.idetect=IDetect(classes,anchors=anchors,ch=[256,512,1024])
s = 256 # 2x min stride 把256带入求得缩放的倍数
self.idetect.stride=torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])
self.idetect.anchors /= self.idetect.stride.view(-1, 1, 1)
# check_anchor_order(self.idetect)
self.stride = self.idetect.stride
self._initialize_biases() # only run once
# Init weights, biases
initialize_weights(self)
def forward(self,x):
x3=self.layer3(self.layer2(self.layer1(self.layer0(x))))
x24=self.layer24(self.layer16(self.layer11(x3)))
x37=self.layer37(self.layer29(x24))
x51=self.layer51(self.layer50(self.layer42(x37)))
x63=self.layer63(self.concat([self.layer54(x37),self.upsample(self.layer52(x51))]))
x75=self.layer75(self.concat([self.layer66(x24),self.upsample(self.layer64(x63))]))
x88=self.layer88(self.concat([self.layer80(x75),x63]))
x101=self.layer101(self.concat([self.layer93(x88),x51]))
x102=self.layer102(x75)
x103=self.layer103(x88)
x104=self.layer104(x101)
out=self.idetect([x102,x103,x104])
return out
def _initialize_biases(self):
for mi,s in zip(self.idetect.m,self.idetect.stride):
b=mi.bias.view(self.idetect.na,-1) # conv.bias(255) to (3,85)
b.data[:,4]+= math.log(8/(640/s)**2)
b.data[:, 5:] += math.log(0.6 / (self.idetect.nc - 0.99)) # cls
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
# device=torch.device('cuda:0')
# anchors=[[12,16, 19,36, 40,28],[36,75, 76,55, 72,146],[142,110, 192,243, 459,401]]
#
# yolov7 = Model(classes=1,anchors=anchors).to(device)
# img = torch.rand( 1, 3, 640, 640).to(device)
# y = yolov7(img)
# print(y[0].shape)
2.各个模块
还是对照网格图
链接: link
# coding=utf-8
import torch
import torch.nn as nn
'''
Author:Don
date:2022/10/17 16:24
desc:
'''
def autopad(k,p=None):
if p is None:
p=k//2 if isinstance(k,int) else [x//2 for x in k]
return p
class Conv(nn.Module):
# ch_in, ch_out, kernel, stride, padding, groups
def __init__(self,c1,c2,k=1,s=1,p=None,g=1,act=True):
super(Conv, self).__init__()
self.conv=nn.Conv2d(c1,c2,k,s,autopad(k,p),groups=g,bias=False)
self.bn=nn.BatchNorm2d(c2)
self.act=nn.SiLU() if act is True else (act if isinstance(act,nn.Module) else nn.Identity())
def forward(self,x):
return self.act(self.bn(self.conv(x)))
def fuseforward(self,x):
return self.act(self.conv(x))
class Concat(nn.Module):
def __init__(self,dimension=1):
super(Concat, self).__init__()
self.d=dimension
def forward(self,x):
return torch.cat(x,self.d)
class MP(nn.Module):
def __init__(self,k=2):
super(MP, self).__init__()
self.m=nn.MaxPool2d(kernel_size=k,stride=k)
def forward(self,x):
return self.m(x)
class MP_2(nn.Module):
def __init__(self,c1):
super(MP_2, self).__init__()
self.conv1=Conv(c1,c1,1,1)
self.conv2=Conv(c1,c1,3,2)
self.m=MP()
self.concat=Concat(1)
def forward(self,x):
y1=self.conv1(self.m(x))
y2=self.conv2(self.conv1(x))
return self.concat([y2,y1])
class MP_1(nn.Module):
def __init__(self,c1):
super(MP_1, self).__init__()
c2=c1//2
self.m=MP()
self.conv1=Conv(c1,c2,1,1)
self.conv2=Conv(c2,c2,3,2)
self.concat=Concat(1)
def forward(self,x):
y1=self.conv1(self.m(x))
y2=self.conv2(self.conv1(x))
return self.concat([y1,y2])
class ELAN(nn.Module):
def __init__(self, c1):
super(ELAN, self).__init__()
c2=c1//2
c3=c1*2
self.conv1 = Conv(c1, c2, 1, 1)
self.conv2 = Conv(c2, c2, 3, 1)
self.conv3 = Conv(c3, c3, 1, 1)
self.concat=Concat(1)
def forward(self,x):
y1=self.conv1(x)
y2=self.conv1(x)
y3=self.conv2(self.conv2(y2))
y4=self.conv2(self.conv2(y3))
return self.conv3(self.concat([y4,y3,y2,y1]))
class ELAN_E(nn.Module):
def __init__(self, c1):
super(ELAN_E, self).__init__()
c2=c1//4
c3=c1
self.conv1 = Conv(c1, c2, 1, 1)
self.conv2 = Conv(c2, c2, 3, 1)
self.conv3 = Conv(c3, c3, 1, 1)
self.concat=Concat(1)
def forward(self,x):
y1=self.conv1(x)
y2=self.conv1(x)
y3=self.conv2(self.conv2(y2))
y4=self.conv2(self.conv2(y3))
return self.conv3(self.concat([y4,y3,y2,y1]))
class ELAN_W(nn.Module):
def __init__(self, c1):
super(ELAN_W, self).__init__()
c2=c1//2
c3=c2//2
c4=2*c2+4*c3
self.conv1 = Conv(c1, c2, 1, 1)
self.conv2 = Conv(c2, c3, 3, 1)
self.conv3 = Conv(c3, c3, 3, 1)
self.conv4 = Conv(c4, c4//4, 1, 1)
self.concat=Concat(1)
def forward(self,x):
y1=self.conv1(x)
y2=self.conv1(x)
y3=self.conv2(y2)
y4=self.conv3(y3)
y5=self.conv3(y4)
y6=self.conv3(y5)
return self.conv4(self.concat([y6,y5,y4,y3,y2,y1]))
class SPPCSPC(nn.Module):
def __init__(self,c1,c2,e=0.5,k=(5,9,13)):
super(SPPCSPC, self).__init__()
c_=int(2*c2*e) #hidden channels
self.cv1=Conv(c1,c_,1,1)
self.cv2=Conv(c1,c_,1,1)
self.cv3=Conv(c_,c_,3,1)
self.cv4=Conv(c_,c_,1,1)
self.m=nn.ModuleList([nn.MaxPool2d(kernel_size=x,stride=1,padding=x//2) for x in k])
self.cv5=Conv(4*c_,c_,1,1)
self.cv6=Conv(c_,c_,3,1)
self.cv7=Conv(2*c_,c2,1,1)
def forward(self,x):
x1=self.cv4(self.cv3(self.cv1(x)))
y1=self.cv6(self.cv5(torch.cat([x1]+[m(x1) for m in self.m],1)))
y2=self.cv2(x)
return self.cv7(torch.cat((y1,y2),dim=1))
class RepConv(nn.Module):
def __init__(self,c1,c2,k=3,s=1,p=None,g=1,act=True,deploy=False):
super(RepConv, self).__init__()
self.deploy=deploy
self.groups=g
self.in_channels=c1
self.out_channels=c2
padding_11=autopad(k,p)-k//2
self.act=nn.SiLU() if act is True else (act if isinstance(act,nn.Module) else nn.Identity())
if deploy:
self.rbr_reparam=nn.Conv2d(c1,c2,k,s,autopad(k,p),groups=g,bias=True)
else:
self.rbr_dense = nn.Sequential(
nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False),
nn.以上是关于yolov7代码梳理，读取json格式数据的主要内容，如果未能解决你的问题，请参考以下文章
Android studio解析含有GPS地理坐标的Json格式数据
如何循环 JSON 数据以获取所有对象的打印百分比值，使用 for each 或 other [重复]
物联网7.物联网通信 - 数据格式(XML ,JSON,MessagePack)