YOLOv5 全面解析教程①：网络结构逐行代码解读

nc: 80  # number of classes 数据集中的类别数depth_multiple: 0.33  # model depth multiple  模型层数因子(用来调整网络的深度)width_multiple: 0.50  # layer channel multiple 模型通道数因子(用来调整网络的宽度)# 如何理解这个depth_multiple和width_multiple呢?它决定的是整个模型中的深度（层数）和宽度（通道数）,具体怎么调整的结合后面的backbone代码解释。
anchors: # 表示作用于当前特征图的Anchor大小为 xxx# 9个anchor，其中P表示特征图的层级，P3/8该层特征图缩放为1/8,是第3层特征  - [10,13, 16,30, 33,23]  # P3/8， 表示[10,13],[16,30], [33,23]3个anchor  - [30,61, 62,45, 59,119]  # P4/16  - [116,90, 156,198, 373,326]  # P5/32

# YOLOv5s v6.0 backbonebackbone:  # [from, number, module, args]  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4   [-1, 3, C3, [128]],   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8   [-1, 6, C3, [256]],   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16   [-1, 9, C3, [512]],   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32   [-1, 3, C3, [1024]],   [-1, 1, SPPF, [1024, 5]],  # 9  ]
# YOLOv5s v6.0 headhead:  [[-1, 1, Conv, [512, 1, 1]],   [-1, 1, nn.Upsample, [None, 2, 'nearest']],   [[-1, 6], 1, Concat, [1]],  # cat backbone P4   [-1, 3, C3, [512, False]],  # 13
   [-1, 1, Conv, [256, 1, 1]],   [-1, 1, nn.Upsample, [None, 2, 'nearest']],   [[-1, 4], 1, Concat, [1]],  # cat backbone P3   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
   [-1, 1, Conv, [256, 3, 2]],   [[-1, 14], 1, Concat, [1]],  # cat head P4   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
   [-1, 1, Conv, [512, 3, 2]],   [[-1, 10], 1, Concat, [1]],  # cat head P5   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)  ]

class Conv(nn.Module):    # Standard convolution    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups        """        @Pargm c1: 输入通道数        @Pargm c2: 输出通道数        @Pargm k : 卷积核大小(kernel_size)        @Pargm s : 卷积步长 (stride)        @Pargm p : 特征图填充宽度 (padding)        @Pargm g : 控制分组，必须整除输入的通道数(保证输入的通道能被正确分组)        """        super().__init__()        # https://oneflow.readthedocs.io/en/master/generated/oneflow.nn.Conv2d.html?highlight=Conv        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 forward_fuse(self, x):        return self.act(self.conv(x))

anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']

n = n_ = max(round(n * gd), 1) if n > 1 else n  # depth gain 暂且将这段代码当作公式(1)

def parse_model(d, ch):  # model_dict, input_channels(3)    """用在下面Model模块中    解析模型文件(字典形式)，并搭建网络结构    这个函数其实主要做的就是: 更新当前层的args（参数）,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 模型文件 字典形式 {dict:7}  [yolov5s.yaml](https://github.com/Oneflow-Inc/one-yolov5/blob/main/models/yolov5s.yaml)中的6个元素 + ch    #Params ch: 记录模型每一层的输出channel 初始ch=[3] 后面会删除    @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]    """    LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10}  {'module':<40}{'arguments':<30}")    # 读取d字典中的anchors和parameters(nc、depth_multiple、width_multiple)    anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']    # na: number of anchors 每一个predict head上的anchor数 = 3    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5) 每一个predict head层的输出channel     # 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中不是-1的层结构序号    # c2: 保存当前层的输出channel    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out    # enumerate() 函数用于将一个可遍历的数据对象(如列表、元组或字符串)组合为一个索引序列，同时列出数据和数据下标，一般用在 for 循环当中。    for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args        m = eval(m) if isinstance(m, str) else m  # eval strings        for j, a in enumerate(args):            # args是一个列表，这一步把列表中的内容取出来            with contextlib.suppress(NameError):                args[j] = eval(a) if isinstance(a, str) else a  # eval strings                # 将深度与深度因子相乘，计算层深度。深度最小为1.         n = n_ = max(round(n * gd), 1) if n > 1 else n  # depth gain                # 如果当前的模块m在本项目定义的模块类型中，就可以处理这个模块        if m in (Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,                 BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x):            # c1: 输入通道数 c2：输出通道数            c1, c2 = ch[f], args[0]             # 该层不是最后一层，则将通道数乘以宽度因子 也就是说，宽度因子作用于除了最后一层之外的所有层            if c2 != no:  # if not output                # make_divisible的作用，使得原始的通道数乘以宽度因子之后取整到8的倍数，这样处理一般是让模型的并行性和推理性能更好。                c2 = make_divisible(c2 * gw, 8)
            # 将前面的运算结果保存在args中，它也就是这个模块最终的输入参数。            args = [c1, c2, *args[1:]]             # 根据每层网络参数的不同，分别处理参数 具体各个类的参数是什么请参考它们的__init__方法这里不再详细解释了            if m in [BottleneckCSP, C3, C3TR, C3Ghost, C3x]:                # 这里的意思就是重复n次，比如conv这个模块重复n次，这个n 是上面算出来的 depth                 args.insert(2, n)  # number of repeats                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        else:            c2 = ch[f]        # 构建整个网络模块 这里就是根据模块的重复次数n以及模块本身和它的参数来构建这个模块和参数对应的Module        m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args)  # module        # 获取模块(module type)具体名例如 models.common.Conv , models.common.C3 , models.common.SPPF 等。        t = str(m)[8:-2].replace('__main__.', '')  #  replace函数作用是字符串"__main__"替换为''，在当前项目没有用到这个替换。        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        """        如果x不是-1，则将其保存在save列表中，表示该层需要保存特征图。        这里 x % i 与 x 等价例如在最后一层 :         f = [17,20,23] , i = 24         y = [ x % i for x in ([f] if isinstance(f, int) else f) if x != -1 ]        print(y) # [17, 20, 23]         # 写成x % i 可能因为：i - 1 = -1 % i (比如 f = [-1]，则 [x % i for x in f] 代表 [11] )        """        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist        layers.append(m_)        if i == 0: # 如果是初次迭代，则新创建一个ch（因为形参ch在创建第一个网络模块时需要用到，所以创建网络模块之后再初始化ch）            ch = []        ch.append(c2)    # 将所有的层封装为nn.Sequential , 对保存的特征图排序    return nn.Sequential(*layers), sorted(save) 

class Model(nn.Module):    # YOLOv5 model    def __init__(self, cfg='[yolov5s.yaml](https://github.com/Oneflow-Inc/one-yolov5/blob/main/models/yolov5s.yaml)', ch=3, nc=None, anchors=None):  # model, input channels, number of classes        super().__init__()        # 如果cfg已经是字典，则直接赋值，否则先加载cfg路径的文件为字典并赋值给self.yaml。        if isinstance(cfg, dict):             self.yaml = cfg  # model dict        else:  # is *.yaml  加载yaml模块            import yaml  # for flow hub             self.yaml_file = Path(cfg).name            with open(cfg, encoding='ascii', errors='ignore') as f:                self.yaml = yaml.safe_load(f)  # model dict  从yaml文件中加载出字典
        # Define model        # ch: 输入通道数。 假如self.yaml有键‘ch’，则将该键对应的值赋给内部变量ch。假如没有‘ch’，则将形参ch赋给内部变量ch        ch = self.yaml['ch'] = self.yaml.get('ch', ch)  # input channels        # 假如yaml中的nc和方法形参中的nc不一致，则覆盖yaml中的nc。        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: # anchors  先验框的配置            LOGGER.info(f'Overriding model.yaml anchors with anchors={anchors}')            self.yaml['anchors'] = round(anchors)  # override yaml value        # 得到模型，以及对应的保存的特征图列表。            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 初始化类名列表，默认为[0,1,2...]                # self.inplace=True  默认True  节省内存        self.inplace = self.yaml.get('inplace', True)
        # Build strides, anchors  确定步长、步长对应的锚框        m = self.model[-1]  # Detect()        if isinstance(m, Detect): # 检验模型的最后一层是Detect模块            s = 256  # 2x min stride            m.inplace = self.inplace            # 计算三个feature map下采样的倍率  [8, 16, 32]            m.stride = flow.tensor([s / x.shape[-2] for x in self.forward(flow.zeros(1, ch, s, s))])  # forward            # 检查anchor顺序与stride顺序是否一致 anchor的顺序应该是从小到大，这里排一下序            check_anchor_order(m)  # must be in pixel-space (not grid-space)            # 对应的anchor进行缩放操作，原因：得到anchor在实际的特征图中的位置，因为加载的原始anchor大小是相对于原图的像素，但是经过卷积池化之后，特征图的长宽变小了。            m.anchors /= m.stride.view(-1, 1, 1)            self.stride = m.stride            self._initialize_biases() # only run once  初始化偏置 
        # Init weights, biases        # 调用oneflow_utils.py下initialize_weights初始化模型权重        initialize_weights(self)        self.info() # 打印模型信息        LOGGER.info('')    # 管理前向传播函数    def forward(self, x, augment=False, profile=False, visualize=False):        if augment:# 是否在测试时也使用数据增强  Test Time Augmentation(TTA)            return self._forward_augment(x)  # augmented inference, None        return self._forward_once(x, profile, visualize)  # single-scale inference, train    # 带数据增强的前向传播    def _forward_augment(self, x):        img_size = x.shape[-2:]  # height, width        s = [1, 0.83, 0.67]  # scales        f = [None, 3, None]  # flips (2-ud, 3-lr)        y = []  # outputs        for si, fi in zip(s, f):            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            yi = self._descale_pred(yi, fi, si, img_size)            y.append(yi)        y = self._clip_augmented(y)  # clip augmented tails        return flow.cat(y, 1), None  # augmented inference, train    # 前向传播具体实现    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 feature_vis: 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，因为后面的特征融合操作要用到这些特征图        y, dt = [], []  # outputs        # 前向推理每一层结构   m.i=index   m.f=from   m.type=类名   m.np=number of params        for m in self.model:            # if not from previous layer   m.f=当前层的输入来自哪一层的输出  s的m.f都是-1            if m.f != -1:  # if not from previous layer                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            if profile:                self._profile_one_layer(m, x, dt)            x = m(x)  # run            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    # 将推理结果恢复到原图图片尺寸(逆操作)    def _descale_pred(self, p, flips, scale, img_size):        # de-scale predictions following augmented inference (inverse operation)        """用在上面的__init__函数上        将推理结果恢复到原图图片尺寸  Test Time Augmentation(TTA)中用到         de-scale predictions following augmented inference (inverse operation)        @params p: 推理结果        @params flips:        @params scale:        @params img_size:        """        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 = flow.cat((x, y, wh, p[..., 4:]), -1)        return p    # 这个是TTA的时候对原图片进行裁剪，也是一种数据增强方式，用在TTA测试的时候。    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    # 打印日志信息  前向推理时间    def _profile_one_layer(self, m, x, dt):        c = isinstance(m, Detect)  # 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")    # initialize biases into Detect(), cf is class frequency    def _initialize_biases(self, cf=None):         # https://arxiv.org/abs/1708.02002 section 3.3        # cf = flow.bincount(flow.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).detach()  # conv.bias(255) to (3,85)            b[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)            b[:, 5:] += math.log(0.6 / (m.nc - 0.999999)) if cf is None else flow.log(cf / cf.sum())  # cls            mi.bias = flow.nn.Parameter(b.view(-1), requires_grad=True)    #  打印模型中最后Detect层的偏置biases信息(也可以任选哪些层biases信息)    def _print_biases(self):        """        打印模型中最后Detect模块里面的卷积层的偏置biases信息(也可以任选哪些层biases信息)        """        m = self.model[-1]  # Detect() module        for mi in m.m:  # from            b = mi.bias.detach().view(m.na, -1).T  # conv.bias(255) to (3,85)            LOGGER.info(                ('%6g Conv2d.bias:' + '%10.3g' * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
    def _print_weights(self):        """        打印模型中Bottleneck层的权重参数weights信息(也可以任选哪些层weights信息)        """        for m in self.model.modules():            if type(m) is Bottleneck:                LOGGER.info('%10.3g' % (m.w.detach().sigmoid() * 2))  # shortcut weights        # fuse()是用来进行conv和bn层合并，为了提速模型推理速度。    def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers        """用在detect.py、val.py        fuse model Conv2d() + BatchNorm2d() layers        调用oneflow_utils.py中的fuse_conv_and_bn函数和common.py中Conv模块的fuseforward函数        """        LOGGER.info('Fusing layers... ')        for m in self.model.modules():            # 如果当前层是卷积层Conv且有bn结构, 那么就调用fuse_conv_and_bn函数讲conv和bn进行融合, 加速推理            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  移除bn remove batchnorm                m.forward = m.forward_fuse  # update forward 更新前向传播 update forward (反向传播不用管, 因为这种推理只用在推理阶段)        self.info()  # 打印conv+bn融合后的模型信息        return self    # 打印模型结构信息 在当前类__init__函数结尾处有调用    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):            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 Detect(nn.Module):    """    Detect模块是用来构建Detect层的，将输入feature map 通过一个卷积操作和公式计算到我们想要的shape, 为后面的计算损失或者NMS后处理作准备    """    stride = None  # strides computed during build    onnx_dynamic = False  # ONNX export parameter    export = False  # export mode
    def __init__(self, nc=80, anchors=(), ch=(), inplace=True):  # detection layer        super().__init__()        #  nc:分类数量        self.nc = nc  # number of classes          #  no:每个anchor的输出数        self.no = nc + 5  # number of outputs per anchor        # nl:预测层数，此次为3        self.nl = len(anchors)  # number of detection layers        #  na:anchors的数量，此次为3        self.na = len(anchors[0]) // 2  # number of anchors        #  grid:格子坐标系，左上角为(1,1),右下角为(input.w/stride,input.h/stride)        self.grid = [flow.zeros(1)] * self.nl  # init grid        self.anchor_grid = [flow.zeros(1)] * self.nl  # init anchor grid        # 写入缓存中，并命名为anchors        self.register_buffer('anchors', flow.tensor(anchors).float().view(self.nl, -1, 2))  # shape(nl,na,2)        # 将输出通过卷积到 self.no * self.na 的通道，达到全连接的作用        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv        self.inplace = inplace  # use inplace ops (e.g. slice assignment)
    def forward(self, x):        z = []  # inference output        for i in range(self.nl):            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)            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
            if not self.training:  # inference                if self.onnx_dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:                    # 向前传播时需要将相对坐标转换到grid绝对坐标系中                    self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)                y = x[i].sigmoid()                if self.inplace:                    y[..., 0:2] = (y[..., 0:2] * 2 + self.grid[i]) * self.stride[i]  # xy                    y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh                else:  # for YOLOv5 on AWS Inferentia https://github.com/ultralytics/yolov5/pull/2953                    xy, wh, conf = y.split((2, 2, self.nc + 1), 4)  # y.tensor_split((2, 4, 5), 4)                      xy = (xy * 2 + self.grid[i]) * self.stride[i]  # xy                    wh = (wh * 2) ** 2 * self.anchor_grid[i]  # wh                    y = flow.cat((xy, wh, conf), 4)                z.append(y.view(bs, -1, self.no))
        return x if self.training else (flow.cat(z, 1),) if self.export else (flow.cat(z, 1), x)        # 相对坐标转换到grid绝对坐标系    def _make_grid(self, nx=20, ny=20, i=0):        d = self.anchors[i].device        t = self.anchors[i].dtype        shape = 1, self.na, ny, nx, 2  # grid shape        y, x = flow.arange(ny, device=d, dtype=t), flow.arange(nx, device=d, dtype=t)               yv, xv = flow.meshgrid(y, x, indexing="ij")        grid = flow.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