yolo.py的主要功能是构建模型。
1、最主要的函数是parse_model,用于解析yaml文件,并根据解析的结果搭建网络。这个函数的注释如下:
def parse_model(d, ch): # model_dict, input_channels(3)
"""
解析模型文件,并搭建网络结构
主要实现功能:更新当前层args,计算c2(当前层的输出channel) => 使用当前层的参数搭建当前层 => 生成layers+save
:params d: model_dict
:params ch: 记录模型每一层的输出channel,初始 ch=[3],后边会删除
:return nn.Sequential(*layers): 网络的每一层的层结构
:return sorted(save): 把所有层结构中from不是-1的值记下,并排序[4,6,10,14,17,20,23]
"""
# Parse a YOLOv5 model.yaml dictionary
LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10} {'module':<40}{'arguments':<30}")
# 可以在yaml文件中指定激活函数,如何使用待定
anchors, nc, gd, gw, act = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple'], d.get('activation')
if act:
Conv.default_act = eval(act) # redefine default activation, i.e. Conv.default_act = nn.SiLU()
LOGGER.info(f"{colorstr('activation:')} {act}") # print
# 三个检测头的参数:3*(20+5) = 75(VOC)
na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors # number of anchors = 3
no = na * (nc + 5) # number of outputs = anchors * (classes + 5)
############################## 开始搭建网络 ###################################
# layers保存每一层的层结构
# save 记录下所有层结构中from不是-1的结构序号
# c2:保存当前层输出的channel
layers, save, c2 = [], [], ch[-1] # layers, savelist, ch out
# from(当前输入来自哪些层) number(当前的层数) module(当前层类别) args(当前层参数)
for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']): # from, number, module, args
# 得到当前层的真实类名 m = Focus -> <class 'models.common.Focus'>
m = eval(m) if isinstance(m, str) else m # eval strings
for j, a in enumerate(args):
# 处理eval函数中出现变量未定义的情况(NameError)
# 如果a是一个字符串类型(str),则将其作为表达式进行求值,得到结果。如果a不是字符串类型,则直接使用a的值。
# 把结果赋值给args[j],以此实现动态地根据字符串表达式来更新args参数列表
with contextlib.suppress(NameError):
args[j] = eval(a) if isinstance(a, str) else a # eval strings
# n当前层数 gd depth_multiple
n = n_ = max(round(n * gd), 1) if n > 1 else n # depth gain
if m in {
Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,
BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x}:
# c1当前层的输出channel c2当前层的输出channel ch 记录所有层的输出channel
c1, c2 = ch[f], args[0]
# 确保c2*gw能够被8整除,如果不能返回一个能够被8整除的最接近于c2*gw的数
# 如果不最后一层的output,就控制宽度,最后一层的channel必须是no
if c2 != no: # if not output
c2 = make_divisible(c2 * gw, 8)
# 在初始args上更新,加入当前层的输入channel
args = [c1, c2, *args[1:]] # [in_channel,out_channel,*args[1:]]
if m in {BottleneckCSP, C3, C3TR, C3Ghost, C3x}:
# 在第二个位置上插入bottleneck的个数n
args.insert(2, n) # number of repeats
n = 1 # 恢复默认值1
elif m is nn.BatchNorm2d:
# 返回上一层的输出channel
args = [ch[f]]
elif m is Concat:
# 把f中的输出累加到这层的channel
c2 = sum(ch[x] for x in f)
# TODO: channel, gw, gd
elif m in {Detect, Segment}:
# 在args中加入三个Detect层的输出channel
args.append([ch[x] for x in f])
if isinstance(args[1], int): # number of anchors # 几乎不执行
args[1] = [list(range(args[1] * 2))] * len(f)
if m is Segment:
args[3] = make_divisible(args[3] * gw, 8)
elif m is Contract: # 几乎不使用
c2 = ch[f] * args[0] ** 2
elif m is Expand: # 几乎不使用
c2 = ch[f] // args[0] ** 2
else:
# Unsample args不变
c2 = ch[f]
# 调用m(类)根据参数args创建当前层的module并赋值给m_,创建数量为n,
m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args) # module
# 打印一些基本信息
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(f'{i:>3}{str(f):>18}{n_:>3}{np:10.0f} {t:<40}{str(args):<30}') # print
# 把所有层结构中from不是-1的值记下[6,4,14,10,17,20,23]
save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist
# 把当前层结构module加入到layers中
layers.append(m_)
if i == 0:
ch = [] # 去除输入channel
# 把当前层输出的channel加入ch
ch.append(c2)
return nn.Sequential(*layers), sorted(save)
parse_model在DetectionModel的__init__函数中调用。
# 调用parse_model
self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # model, savelist
self.names = [str(i) for i in range(self.yaml['nc'])] # default names
self.inplace = self.yaml.get('inplace', True)
2、Detect类用于构建最后的detect层,在parse_model函数中调用
# 构建Detect层,把feature map通过一个卷积操作和公式计算到需要的shape,为后边计算loss和NMS做准备。
class Detect(nn.Module):
# YOLOv5 Detect head for detection models
stride = None # strides computed during build
dynamic = False # force grid reconstruction
export = False # export mode
def __init__(self, nc=80, anchors=(), ch=(), inplace=True): # detection layer
"""
detection layer相当于yolov3的YOLOLayer层
:params nc: number of class
:params anchors:传入3个feature map上的所有anchor的大小(P3,P4,P5)
:params ch:[128,256,512] 3个输出feanture map的chaannel
"""
super().__init__()
self.nc = nc # number of classes
self.no = nc + 5 # number of outputs per anchor
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors 每个feature map的anchor个数
self.grid = [torch.empty(0) for _ in range(self.nl)] # init grid
self.anchor_grid = [torch.empty(0) for _ in range(self.nl)] # init anchor grid
# 模型中需要保存的参数有两种:一种是需要使用optimizer更新的,一种是不需要被更新的称为buffer
# buffer的参数更新在forward,而optim.step只能更新nn.parameter类型的参数
# anchor.shape = shape(nl,na,2)
self.register_buffer('anchors', torch.tensor(anchors).float().view(self.nl, -1, 2)) # shape(nl,na,2)
# 对每个输出feature map都调用一次conv1*1
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
# 默认为True,默认不使用AWS Inferentia加速
self.inplace = inplace # use inplace ops (e.g. slice assignment)
def forward(self, x):
"""
:return train:一个tensor list存放三个元素[bs, anchor_num, grid_w, grid_h, xywh+classes]
以VOC为例,[1, 3, 80, 80, 25] [1, 3, 40, 40, 25] [1, 3, 20, 20, 25]
inference:0 preds [1, 19200+4800+1200, 25] = [bs, anchor_num*grid_w*grid_h, xywh+classes]
1 train_out :一个tensor list存放三个元素[bs, anchor_num, grid_w, grid_h, xywh+classes]
[1, 3, 80, 80, 25] [1, 3, 40, 40, 25] [1, 3, 20, 20, 25]
"""
z = [] # inference output
for i in range(self.nl): # 对3个feature map分别进行处理
x[i] = self.m[i](x[i]) # conv
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
# [bs, 75, 80, 80] to [bs, 3, 25, 80, 80] to [bs, 3, 80, 80, 25]
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
# inference 返回的不是归一化后的网格的偏移量,需要加上网格的位置,得到最终的预测坐标,再送入NMS
# 构建网络就是为了记录每个grid的网格坐标,方便后边使用
# dynamic默认为false
if self.dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)
if isinstance(self, Segment): # (boxes + masks)
xy, wh, conf, mask = x[i].split((2, 2, self.nc + 1, self.no - self.nc - 5), 4)
xy = (xy.sigmoid() * 2 + self.grid[i]) * self.stride[i] # xy
wh = (wh.sigmoid() * 2) ** 2 * self.anchor_grid[i] # wh
y = torch.cat((xy, wh, conf.sigmoid(), mask), 4)
else: # Detect (boxes only)
xy, wh, conf = x[i].sigmoid().split((2, 2, self.nc + 1), 4)
xy = (xy * 2 + self.grid[i]) * self.stride[i] # xy
wh = (wh * 2) ** 2 * self.anchor_grid[i] # wh
y = torch.cat((xy, wh, conf), 4)
# z是一个teensor list,三个元素,分别是[1,19200,25] [1, 4800, 25], [1, 1200, 25]
z.append(y.view(bs, self.na * nx * ny, self.no))
return x if self.training else (torch.cat(z, 1), ) if self.export else (torch.cat(z, 1), x)
def _make_grid(self, nx=20, ny=20, i=0, torch_1_10=check_version(torch.__version__, '1.10.0')):
"""
生成网格和锚框的张量,网格形状由ny和nx居多,锚框形状由self.anchors[i]和self.stride[i]决定
:params nx: 网格宽度
:params ny: 网格高度
:params i: 锚框索引
:params torch_1_10:判断torch版本是否大于1.10.0
:return grid:
:return anchor_grid:
"""
d = self.anchors[i].device # 锚框的device
t = self.anchors[i].dtype # 锚框的数据类型
# self.na 每个feature map的anchor的个数
shape = 1, self.na, ny, nx, 2 # grid shape
# y是长度为ny的张量,x是长度为nx的张量,y和x分别表示网格的纵坐标和横坐标
y, x = torch.arange(ny, device=d, dtype=t), torch.arange(nx, device=d, dtype=t)
# 如果torch_1_10为True,表示torch版本大于等于1.10.0,代码使用torch.meshgrid函数生成网格坐标,采用'ij'索引方式。
# 否则,代码使用torch.meshgrid函数生成网格坐标,采用默认的索引方式
yv, xv = torch.meshgrid(y, x, indexing='ij') if torch_1_10 else torch.meshgrid(y, x) # torch>=0.7 compatibility
# [ny, nx, 2] to [1, self.na, ny, nx, 2] - 0.5 # 以此得到网格的偏移量
grid = torch.stack((xv, yv), 2).expand(shape) - 0.5 # add grid offset, i.e. y = 2.0 * x - 0.5
anchor_grid = (self.anchors[i] * self.stride[i]).view((1, self.na, 1, 1, 2)).expand(shape)
return grid, anchor_grid
3、DetectionModel类继承自BaseModel类,用于构建模型,使用时定义了一个全局变量model指向这个类,在其他文件中使用时直接调用model。
class BaseModel(nn.Module):
# YOLOv5 base model
def forward(self, x, profile=False, visualize=False):
return self._forward_once(x, profile, visualize) # single-scale inference, train
def _forward_once(self, x, profile=False, visualize=False):
"""
:params x: 输入图像
:params profile: True 可以做一些性能评估
:params visualize: True 可以做一些特征可视化
"""
# y: 存放着self.save = True 的每一层的输出,因为后边层结构concat要用到
# dt:在profile中做性能评估时使用
y, dt = [], [] # outputs
# 前向推理每一层结构
# m.i = index m.f = from m.type = 类名 m.mp = number of parms
for m in self.model:
# 4个concat操作和1个detect操作
if m.f != -1: # if not from previous layer
# concat: m.f=[-1,6] x就有两个元素,一个是上一层的输出,另一个是index层的输出
x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f] # from earlier layers
# 打印日志信息,FLOPS、time
if profile:
self._profile_one_layer(m, x, dt)
x = m(x) # run
# 存放着self.save的每一层的输出
y.append(x if m.i in self.save else None) # save output
if visualize:
feature_visualization(x, m.type, m.i, save_dir=visualize)
return x
def _profile_one_layer(self, m, x, dt):
c = m == self.model[-1] # is final layer, copy input as inplace fix
o = thop.profile(m, inputs=(x.copy() if c else x, ), verbose=False)[0] / 1E9 * 2 if thop else 0 # FLOPs
t = time_sync()
for _ in range(10):
m(x.copy() if c else x)
dt.append((time_sync() - t) * 100)
if m == self.model[0]:
LOGGER.info(f"{'time (ms)':>10s} {'GFLOPs':>10s} {'params':>10s} module")
LOGGER.info(f'{dt[-1]:10.2f} {o:10.2f} {m.np:10.0f} {m.type}')
if c:
LOGGER.info(f"{sum(dt):10.2f} {'-':>10s} {'-':>10s} Total")
def fuse(self): # fuse model Conv2d() + BatchNorm2d() layers
"""
用在detect.py和val.py中
fuse model conv2d() + batch norm
"""
LOGGER.info('Fusing layers... ')
for m in self.model.modules():
if isinstance(m, (Conv, DWConv)) and hasattr(m, 'bn'):
m.conv = fuse_conv_and_bn(m.conv, m.bn) # update conv
delattr(m, 'bn') # remove batchnorm
m.forward = m.forward_fuse # update forward
self.info()
return self
def info(self, verbose=False, img_size=640): # print model information
model_info(self, verbose, img_size)
def _apply(self, fn):
# Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
self = super()._apply(fn)
m = self.model[-1] # Detect()
if isinstance(m, (Detect, Segment)):
m.stride = fn(m.stride)
m.grid = list(map(fn, m.grid))
if isinstance(m.anchor_grid, list):
m.anchor_grid = list(map(fn, m.anchor_grid))
return self
class DetectionModel(BaseModel):
# YOLOv5 detection model
def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None, anchors=None): # model, input channels, number of classes
"""
params cfg: 配置文件
params ch: input channel
param nc: number of classes
params anchor: 一般是none
"""
super().__init__()
if isinstance(cfg, dict):
self.yaml = cfg # model dict
else: # is *.yaml
import yaml # for torch hub
self.yaml_file = Path(cfg).name
with open(cfg, encoding='ascii', errors='ignore') as f:
self.yaml = yaml.safe_load(f) # model dict
# Define model
ch = self.yaml['ch'] = self.yaml.get('ch', ch) # input channels
if nc and nc != self.yaml['nc']:
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) # override yaml value
# 调用parse_model 创建网络模型
# self.model 初始化的整个网络结构
# self.save 所有层结构中from不等于-1的序号
self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # model, savelist
# default class names ['0', '1', '2', '3', '4'.......]
self.names = [str(i) for i in range(self.yaml['nc'])] # default names
self.inplace = self.yaml.get('inplace', True)
# Build strides, anchors
# 获取Detect 模块的stride(相对于输入图像的下采样率)和anchors在当前Detect输出的feature map的尺度
m = self.model[-1] # Detect()
if isinstance(m, (Detect, Segment)):
s = 256 # 2x min stride
m.inplace = self.inplace
forward = lambda x: self.forward(x)[0] if isinstance(m, Segment) else self.forward(x)
# 计算三个feature map的下采样率
m.stride = torch.tensor([s / x.shape[-2] for x in forward(torch.zeros(1, ch, s, s))]) # forward
# 检查anchor顺序是否与stride顺序是否一致
check_anchor_order(m)
# 求出相对于当前feature map的anchor大小 [10,13]/8=[1.25,1.625]
m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
self._initialize_biases() # only run once
# Init weights, biases
initialize_weights(self) # 初始化模型权重
self.info()
LOGGER.info('')
def forward(self, x, augment=False, profile=False, visualize=False):
# 是否在测试时使用Test Time Augmentation(TTA)
if augment:
return self._forward_augment(x) # augmented inference, None
# 默认执行前向推理
# single-scale inference, train
# _forward_once在BaseModel中实现
return self._forward_once(x, profile, visualize) # single-scale inference, train
def _forward_augment(self, x):
"""Test Time Augmentation(TTA) """
img_size = x.shape[-2:] # height, width
s = [1, 0.83, 0.67] # scales
f = [None, 3, None] # flips (2-ud上下, 3-lr左右)
y = [] # outputs
for si, fi in zip(s, f):
# scale_img缩放图片尺寸
xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
yi = self._forward_once(xi)[0] # forward
# cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1]) # save
# descale
yi = self._descale_pred(yi, fi, si, img_size)
y.append(yi)
y = self._clip_augmented(y) # clip augmented tails
return torch.cat(y, 1), None # augmented inference, train
def _descale_pred(self, p, flips, scale, img_size):
# de-scale predictions following augmented inference (inverse operation)
if self.inplace:
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
else:
x, y, wh = p[..., 0:1] / scale, p[..., 1:2] / scale, p[..., 2:4] / scale # de-scale
if flips == 2:
y = img_size[0] - y # de-flip ud
elif flips == 3:
x = img_size[1] - x # de-flip lr
p = torch.cat((x, y, wh, p[..., 4:]), -1)
return p
def _clip_augmented(self, y):
# Clip YOLOv5 augmented inference tails
nl = self.model[-1].nl # number of detection layers (P3-P5)
g = sum(4 ** x for x in range(nl)) # grid points
e = 1 # exclude layer count
i = (y[0].shape[1] // g) * sum(4 ** x for x in range(e)) # indices
y[0] = y[0][:, :-i] # large
i = (y[-1].shape[1] // g) * sum(4 ** (nl - 1 - x) for x in range(e)) # indices
y[-1] = y[-1][:, i:] # small
return y
def _initialize_biases(self, cf=None): # initialize biases into Detect(), cf is class frequency
# https://arxiv.org/abs/1708.02002 section 3.3
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
m = self.model[-1] # Detect() module
for mi, s in zip(m.m, m.stride): # from
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
b.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
b.data[:, 5:5 + m.nc] += math.log(0.6 / (m.nc - 0.99999)) if cf is None else torch.log(cf / cf.sum()) # cls
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
使用语句
Model = DetectionModel # retain YOLOv5 'Model' class for backwards compatibility