抄写yolov5---1模型
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，主要介绍了抄写yolov5---1模型相关的知识，希望对你有一定的参考价值。
yolov5代码梳理
1.yolo.py
import torch.nn as nn
import yaml
import argparse
import logging
from pathlib import Path
import sys
from copy import deepcopy
from layers import *
from utils.general import *
sys.path.append(Path(__file__).parent.parent.absolute().__str__()) # to run '$ python *.py' files in subdirectories
logger = logging.getLogger(__name__)
class Detect(nn.Module):
stride = None
def __init__(self, nc=80, anchors=(), ch=(), inplace=True): # 类别，anchros，输入的层
super(Detect, self).__init__()
self.nc = nc
self.no = nc + 5 # 每一个annchor 输出的个数 85
self.nl = len(anchors) # 检测的层 3
self.na = len(anchors[0]) // 2 # 每一个网络有几个anchor 3
self.grid = [torch.zeros(1)] * self.nl # 初始化网格 [tensor([0.]), tensor([0.]), tensor([0.])]
a = torch.tensor(anchors).float().view(self.nl, -1, 2) # 3,-1,2
self.register_buffer('anchors', a) # 3,3,2
self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2)) # 3,1,-1,1,1,2
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)
self.inplace = inplace
def forward(self, x):
z = []
for i in range(self.nl): # 3
x[i] = self.m[i](x[i]) # 把3个输出层 转化为255通道 bs,255,20,20
bs, _, ny, nx = x[i].shape # batch_size,255,20,20 to batch_size,3,20,20,85
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # 推理阶段 todo
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device) # 1,1,20,20,2
y = x[i].sigmoid(x)
if self.inplace:
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # x,y 缩小的倍数
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
z.append(y.view(bs, -1, self.no))
return x if self.training else (torch.cat(z, 1), x)
@staticmethod
def _make_grid(nx=20, ny=20):
yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
def parse_model(d, ch): # 修改后的配置文件‘yolov5.yaml’，输人通道[3]
logger.info('\\n%3s%18s%3s%10s %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d[
'width_multiple'] # anchor ,类别（80）， 模型深度缩减比例（ 0.33），模型宽度缩减比例（0.50）
na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors # 每一个检测层配置的anchor的个数，这里是3
no = na * (nc + 5) # 输出大小，每一个anchor要输出类别加上xywh和置信度， 3*（80+5）
# no=na*(nc+5+180) # 输出大小，每一个anchor要输出类别加上xywh和置信度和角度分类， 3*（80+5+180）
layers, save, c2 = [], [], ch[-1] # 层，保存的数据，输出
for i, (f, n, m, args) in enumerate(
d['backbone'] + d['head']): # 加载配置文件中backone和head部分 ，[来自那一层，层个数，层名字，其他参数（卷积核大小和步长）]
m = eval(m) if isinstance(m, str) else m # 输出str类型的层名字
for j, a in enumerate(args): # 输出str格式的参数
try:
args[j] = eval(a) if isinstance(a, str) else a #
except:
pass
n = max(round(n * gd), 1) if n > 1 else n # 修改网络的深度，如果大于1 就乘以相应缩减比例
if m in [CBA, GhostConv, Bottleneck, GhostBottleneck, SPP, DWConv, MixConv2d, Focus, CrossConv, BottleneckCSP,
C3, C3TR, SELayer]:
c1, c2 = ch[f], args[0]
if c2 != no: # if not output
c2 = make_divisible(c2 * gw, 8) # 保证是8的倍数
args = [c1, c2, *args[1:]]
if m in [BottleneckCSP, C3, C3TR]:
args.insert(2, n) # number of repeats 多个层
n = 1
elif m is nn.BatchNorm2d:
args = [ch[f]]
elif m is Concat:
c2 = sum([ch[x] for x in f])
elif m is Detect:
args.append([ch[x] for x in f])
if isinstance(args[1], int): # number of anchors
args[1] = [list(range(args[1] * 2))] * len(f)
elif m is Contract:
c2 = ch[f] * args[0] ** 2
elif m is Expand:
c2 = ch[f] // args[0] ** 2
elif m is SELayer:
channel = args[0]
channel = make_divisible(channel * gw, 8) if channel != no else channel
args = [channel]
else:
c2 = ch[f]
m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args) # module
t = str(m)[8:-2].replace('__main__.', '') # module type
np = sum([x.numel() for x in m_.parameters()]) # number params
m_.i, m_.f, m_.type, m_.np = i, f, t, np # attach index, 'from' index, type, number params
logger.info('%3s%18s%3s%10.0f %-40s%-30s' % (i, f, n, np, t, args)) # print
save.extend(
x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist [6,4,14,10,17,20,23]
layers.append(m_)
if i == 0:
ch = []
ch.append(c2)
return nn.Sequential(*layers), sorted(save)
def initialize_weights(model):
for m in model.modules():
t=type(m)
if t is nn.Conv2d:
pass
elif t is nn.BatchNorm2d:
m.eps=1e-3
m.momentum=0.03
elif t in [nn.Hardswish,nn.LeakyReLU,nn.ReLU,nn.ReLU6]:
m.inplace=True
class Model(nn.Module):
def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None, anchors=None): # 模型、输人、类别、anchors
super(Model, self).__init__()
if isinstance(cfg, dict): # 如果是字典类型表示已经加载过了,否则加载模型配置文件
self.yaml = cfg
else:
with open(cfg) as f:
self.yaml = yaml.safe_load(f)
# 定义模型
ch = self.yaml['ch'] = self.yaml.get('ch', ch) # input channels#图片输入通道
if nc and nc != self.yaml['nc']: # 如果设置模型的类别和配置文件不一样，修改配置文件
logger.info(f"Overriding model.yaml nc={self.yaml['nc']} with nc={nc}")
self.yaml['nc'] = nc # override yaml value
if anchors:
logger.info(f'Overriding model.yaml anchors with anchors={anchors}')
self.yaml['anchors'] = round(anchors) # 修改配置文件的anchor
self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # 根据配置文件生成模型和需要保存的层数
self.names = [str(i) for i in range(self.yaml['nc'])] # 类名
self.inplace = self.yaml.get('inplace', True)
m = self.model[-1] # detect
if isinstance(m, Detect):
s = 128
m.inplace = self.inplace
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # 同过前向计算 ，得到缩小比例
m.anchors /= m.stride.view(-1, 1, 1) # 缩小相应比例
check_anchor_order(m) # 检测anchor排序
self.stride = m.stride
self._initialize_biases()
initialize_weights(self) #初始化eights, biases
self.info()
def forward(self, x, augment=False):
if augment: # 图像增强
return self.forward_augment(x)
else:
return self.forward_once(x)
def forward_augment(self,x):
img_size=x.shape[-2:] #h,w
s=[1,0.83,0.67] #缩放
f=[None,3,None] #反转
y=[]
for si,fi in zip(s,f):
xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max())) #对输入图像做图像缩放
yi=self.forward_once(xi)[0]
yi = self._descale_pred(yi, fi, si, img_size) #对输出结果做缩放
y.append(yi)
return torch.cat(y,1),None
def forward_once(self, x):
y, dt = [], []
for m in self.model:
if m.f != -1: # 不是来源前一个层，多个层相加
x = [x if j == -1 else y[j] for j in m.f]
x = m(x)
y.append(x if m.i in self.save else None) # 把[6,4,14,10,17,20,23] 层的输出保存起来
return x
def info(self, verbose=False, img_size=640): # print model information
model_info(self, verbose, img_size)
def _descale_pred(self,p,flips,scale,img_size): #输出，反转，缩放，输入的大小
p[..., :4] /= scale # de-scale
if flips == 2:
p[..., 1] = img_size[0] - p[..., 1] # de-flip ud
elif flips == 3:
p[..., 0] = img_size[1] - p[..., 0] # de-flip lr
return p
def _initialize_biases(self,cf=None): # 初始化检测层的偏执，cf是类别概率
m=self.model[-1]
for mi ,s in zip(m.m,m.stride):
b=mi.bias.view(m.na,-1) #bias 255 to 3,85
b.data[:, 4] += math.log(8 / (640 <以上是关于抄写yolov5---1模型的主要内容，如果未能解决你的问题，请参考以下文章
使用片段时 Intellij 无法正确识别 Thymeleaf 模型变量