Yolov5代码解析(输入端、BackBone、Neck、输出端))

def mixup(im, labels, im2, labels2):
    # Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
    r = np.random.beta(32.0, 32.0)  # mixup ratio, alpha=beta=32.0
    im = (im * r + im2 * (1 - r)).astype(np.uint8)  # 混合图像
    labels = np.concatenate((labels, labels2), 0)  # 标签直接concate更加简单
    return im, labels

def cutout(im, labels, p=0.5):
    # Applies image cutout augmentation https://arxiv.org/abs/1708.04552
    if random.random()  0.03:
                box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
                ioa = bbox_ioa(box, labels[:, 1:5])  # intersection over area
                labels = labels[ioa 60% obscured labels

    return labels

    def load_mosaic(self, index):
        # YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
        labels4, segments4 = [], []
        s = self.img_size
        yc, xc = (int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border)  # mosaic center x, y　有范围限制，左右留下mosaic_border大小边界
        indices = [index] + random.choices(self.indices, k=3)  # 在所有图片中随机选择三张
        random.shuffle(indices)
        for i, index in enumerate(indices):
            # Load image
            img, _, (h, w) = self.load_image(index)

            # place img in img4
            if i == 0:  # top left
                img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles (np.full用固定值填充2s*2s的大图）
                x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc  # xmin, ymin, xmax, ymax (large image)
                x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h  # xmin, ymin, xmax, ymax (small image)
            elif i == 1:  # top right
                x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
                x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
            elif i == 2:  # bottom left
                x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
                x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
            elif i == 3:  # bottom right
                x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)  # 在大图中每张小图的位置
                x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)  # 每张小图中对应大小的区域

            img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax] 将小图拷贝到大图中对应位置
            padw = x1a - x1b  # 小图的左上角点相对于大图左上角点的偏移(padw, padh)，用来计算mosaic增强后的标签框的位置
            padh = y1a - y1b

            # Labels
            labels, segments = self.labels[index].copy(), self.segments[index].copy()
            if labels.size:
                labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padw, padh)  # normalized xywh to pixel xyxy format
                segments = [xyn2xy(x, w, h, padw, padh) for x in segments]
            labels4.append(labels)
            segments4.extend(segments)

        # Concat/clip labels
        labels4 = np.concatenate(labels4, 0)
        for x in (labels4[:, 1:], *segments4):
            np.clip(x, 0, 2 * s, out=x)  # clip when using random_perspective()
        # img4, labels4 = replicate(img4, labels4)  # replicate

        # Augment
        img4, labels4, segments4 = copy_paste(img4, labels4, segments4, p=self.hyp['copy_paste'])
        img4, labels4 = random_perspective(img4,
                                           labels4,
                                           segments4,
                                           degrees=self.hyp['degrees'],
                                           translate=self.hyp['translate'],
                                           scale=self.hyp['scale'],
                                           shear=self.hyp['shear'],
                                           perspective=self.hyp['perspective'],
                                           border=self.mosaic_border)  # border to remove

        return img4, labels4

    def load_mosaic9(self, index):
        # YOLOv5 9-mosaic loader. Loads 1 image + 8 random images into a 9-image mosaic
        labels9, segments9 = [], []
        s = self.img_size
        indices = [index] + random.choices(self.indices, k=8)  # 8 additional image indices
        random.shuffle(indices)
        hp, wp = -1, -1  # height, width previous
        for i, index in enumerate(indices):
            # Load image
            img, _, (h, w) = self.load_image(index)

            # place img in img9
            if i == 0:  # center
                img9 = np.full((s * 3, s * 3, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
                h0, w0 = h, w
                c = s, s, s + w, s + h  # xmin, ymin, xmax, ymax (base) coordinates
            elif i == 1:  # top
                c = s, s - h, s + w, s
            elif i == 2:  # top right
                c = s + wp, s - h, s + wp + w, s
            elif i == 3:  # right
                c = s + w0, s, s + w0 + w, s + h
            elif i == 4:  # bottom right
                c = s + w0, s + hp, s + w0 + w, s + hp + h
            elif i == 5:  # bottom
                c = s + w0 - w, s + h0, s + w0, s + h0 + h
            elif i == 6:  # bottom left
                c = s + w0 - wp - w, s + h0, s + w0 - wp, s + h0 + h
            elif i == 7:  # left
                c = s - w, s + h0 - h, s, s + h0
            elif i == 8:  # top left
                c = s - w, s + h0 - hp - h, s, s + h0 - hp

            padx, pady = c[:2]
            x1, y1, x2, y2 = (max(x, 0) for x in c)  # allocate coords

            # Labels
            labels, segments = self.labels[index].copy(), self.segments[index].copy()
            if labels.size:
                labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padx, pady)  # normalized xywh to pixel xyxy format
                segments = [xyn2xy(x, w, h, padx, pady) for x in segments]
            labels9.append(labels)
            segments9.extend(segments)

            # Image
            img9[y1:y2, x1:x2] = img[y1 - pady:, x1 - padx:]  # img9[ymin:ymax, xmin:xmax]
            hp, wp = h, w  # height, width previous

        # Offset
        yc, xc = (int(random.uniform(0, s)) for _ in self.mosaic_border)  # mosaic center x, y
        img9 = img9[yc:yc + 2 * s, xc:xc + 2 * s]

        # Concat/clip labels
        labels9 = np.concatenate(labels9, 0)
        labels9[:, [1, 3]] -= xc
        labels9[:, [2, 4]] -= yc
        c = np.array([xc, yc])  # centers
        segments9 = [x - c for x in segments9]

        for x in (labels9[:, 1:], *segments9):
            np.clip(x, 0, 2 * s, out=x)  # clip when using random_perspective()
        # img9, labels9 = replicate(img9, labels9)  # replicate

        # Augment
        img9, labels9 = random_perspective(img9,
                                           labels9,
                                           segments9,
                                           degrees=self.hyp['degrees'],
                                           translate=self.hyp['translate'],
                                           scale=self.hyp['scale'],
                                           shear=self.hyp['shear'],
                                           perspective=self.hyp['perspective'],
                                           border=self.mosaic_border)  # border to remove

        return img9, labels9

mosaic = self.mosaic and random.random() 

def copy_paste(im, labels, segments, p=0.5):
    # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
    n = len(segments)
    if p and n:
        h, w, c = im.shape  # height, width, channels
        im_new = np.zeros(im.shape, np.uint8)
        for j in random.sample(range(n), k=round(p * n)):
            l, s = labels[j], segments[j]
            box = w - l[3], l[2], w - l[1], l[4]
            ioa = bbox_ioa(box, labels[:, 1:5])  # intersection over area
            if (ioa  0  # pixels to replace
        # i[:, :] = result.max(2).reshape(h, w, 1)  # act over ch
        im[i] = result[i]  # cv2.imwrite('debug.jpg', im)  # debug

    return im, labels, segments

def random_perspective(im,
                       targets=(),
                       segments=(),
                       degrees=10,
                       translate=.1,
                       scale=.1,
                       shear=10,
                       perspective=0.0,
                       border=(0, 0)):
    # torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
    # targets = [cls, xyxy]

    height = im.shape[0] + border[0] * 2  # shape(h,w,c)
    width = im.shape[1] + border[1] * 2

    # Center
    C = np.eye(3)
    C[0, 2] = -im.shape[1] / 2  # x translation (pixels)
    C[1, 2] = -im.shape[0] / 2  # y translation (pixels)

    # Perspective
    P = np.eye(3)
    P[2, 0] = random.uniform(-perspective, perspective)  # x perspective (about y)
    P[2, 1] = random.uniform(-perspective, perspective)  # y perspective (about x)

    # Rotation and Scale
    R = np.eye(3)
    a = random.uniform(-degrees, degrees)
    # a += random.choice([-180, -90, 0, 90])  # add 90deg rotations to small rotations
    s = random.uniform(1 - scale, 1 + scale)
    # s = 2 ** random.uniform(-scale, scale)
    R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)

    # Shear
    S = np.eye(3)
    S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # x shear (deg)
    S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # y shear (deg)

    # Translation
    T = np.eye(3)
    T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width  # x translation (pixels)
    T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height  # y translation (pixels)

    # Combined rotation matrix
    M = T @ S @ R @ P @ C  # order of operations (right to left) is IMPORTANT
    if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any():  # image changed
        if perspective:
            im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))# 透视变换
        else:  # affine
            im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))# 仿射变换

    # Visualize
    # import matplotlib.pyplot as plt
    # ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
    # ax[0].imshow(im[:, :, ::-1])  # base
    # ax[1].imshow(im2[:, :, ::-1])  # warped
    
    # Transform label coordinates
    n = len(targets)
    if n:
        use_segments = any(x.any() for x in segments)
        new = np.zeros((n, 4))
        if use_segments:  # warp segments
            segments = resample_segments(segments)  # upsample
            for i, segment in enumerate(segments):
                xy = np.ones((len(segment), 3))
                xy[:, :2] = segment
                xy = xy @ M.T  # transform
                xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]  # perspective rescale or affine

                # clip
                new[i] = segment2box(xy, width, height)

        else:  # warp boxes
            xy = np.ones((n * 4, 3))
            xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2)  # x1y1, x2y2, x1y2, x2y1
            xy = xy @ M.T  # transform
            xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8)  # perspective rescale or affine

            # create new boxes
            x = xy[:, [0, 2, 4, 6]]
            y = xy[:, [1, 3, 5, 7]]
            new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T

            # clip
            new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
            new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)

        # filter candidates 对label进行面积，长宽和长宽比筛选
        i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
        targets = targets[i]
        targets[:, 1:5] = new[i]

    return im, targets

def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
    # HSV color-space augmentation
    if hgain or sgain or vgain:
        r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1  # random gains
        hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV)) #由bgr转为hsv后分离三通道
        dtype = im.dtype  # uint8
        # 创建3个通道的查找表，将通过查找表将原值映射为新值
        x = np.arange(0, 256, dtype=r.dtype)
        lut_hue = ((x * r[0]) % 180).astype(dtype)  # opencv中hue值的范围0~180
        lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
        lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
        # H，S，V三个通道将原值映射至随机增减后的值，再合并
        im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
        cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im)  # no return needed

# Flip up-down
if random.random() 

class Albumentations:
    # YOLOv5 Albumentations class (optional, only used if package is installed)
    def __init__(self):
        self.transform = None
        try:
            import albumentations as A
            check_version(A.__version__, '1.0.3', hard=True)  # version requirement

            T = [
                A.Blur(p=0.01),                       # 随机模糊
                A.MedianBlur(p=0.01),                 # 中值滤波器模糊输入图像
                A.ToGray(p=0.01),                     # 将输入的 RGB 图像转换为灰度
                A.CLAHE(p=0.01),                      # 自适应直方图均衡
                A.RandomBrightnessContrast(p=0.0),    # 随机改变输入图像的亮度和对比度
                A.RandomGamma(p=0.0),                 # 随机伽马变换
                A.ImageCompression(quality_lower=75, p=0.0),  # 减少图像的 Jpeg、WebP 压缩

                # 可加
                A.GaussianBlur(p=0.15),               # 高斯滤波器模糊
                A.GaussNoise(p=0.15),                 # 高斯噪声应用于输入图像
                A.FancyPCA(p=0.25),                   # PCA来找出R/G/B这三维的主成分，然后随机增加图像像素强度（AlexNet）
            ]
            self.transform = A.Compose(T, bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))

            LOGGER.info(colorstr('albumentations: ') + ', '.join(f'{x}' for x in self.transform.transforms if x.p))
        except ImportError:  # package not installed, skip
            pass
        except Exception as e:
            LOGGER.info(colorstr('albumentations: ') + f'{e}')

    def __call__(self, im, labels, p=1.0):
        if self.transform and random.random() 

def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
    """在check_anchors中调用
    使用K-means + 遗传算法 算出更符合当前数据集的anchors
    Creates kmeans-evolved anchors from training dataset
    :params path: 数据集的路径/数据集本身
    :params n: anchors 的个数
    :params img_size: 数据集图片约定的大小
    :params thr: 阈值 由 hyp['anchor_t'] 参数控制
    :params gen: 遗传算法进化迭代的次数(突变 + 选择)
    :params verbose: 是否打印所有的进化(成功的)结果 默认传入是False, 只打印最佳的进化结果
    :return k: K-means + 遗传算法进化后的anchors
    """
    from scipy.cluster.vq import kmeans


    # 注意一下下面的thr不是传入的thr，而是1/thr, 所以在计算指标这方面还是和check_anchor一样
    thr = 1. / thr  # 0.25
    prefix = colorstr('autoanchor: ')

    def metric(k, wh):  # compute metrics
        """用于 print_results 函数和 anchor_fitness 函数
        计算ratio metric: 整个数据集的  ground truth 框与 anchor 对应宽比和高比即:gt_w/k_w,gt_h/k_h + x + best_x  用于后续计算BPR+aat
        注意我们这里选择的metric是 ground truth 框与anchor对应宽比和高比 而不是常用的iou 这点也与nms的筛选条件对应 是yolov5中使用的新方法
        :params k: anchor框
        :params wh: 整个数据集的 wh [N, 2]
        :return x: [N, 9] N 个 ground truth 框与所有 anchor 框的宽比或高比(两者之中较小者)
        :return x.max(1)[0]: [N] N个 ground truth 框与所有 anchor 框中的最大宽比或高比(两者之中较小者)
        """
        # [N, 1, 2] / [1, 9, 2] = [N, 9, 2]  N个gt_wh和9个anchor的k_wh宽比和高比
        # 两者的重合程度越高 就越趋近于1 远离1(1)重合程度都越低
        r = wh[:, None] / k[None]
        # r=gt_height/anchor_height  gt_width / anchor_width  有可能大于1，也可能小于等于1
        # flow.min(r, 1. / r): [N, 9, 2] 将所有的宽比和高比统一到  thr).float()).mean()=适应度计算公式 [1] 注意和BPR有区别 这里是自定义的一种适应度公式
                返回的是输入此时anchor k 对应的适应度
        """
        _, best = metric(flow.tensor(k, dtype=flow.float32), wh)
        return (best * (best > thr).float()).mean()  # fitness

    def print_results(k):
        """用于 kmean_anchors 函数中打印K-means计算相关信息
        计算BPR、aat=>打印信息: 阈值+BPR+aat  anchor个数+图片大小+metric_all+best_mean+past_mean+Kmeans聚类出来的anchor框(四舍五入)
        :params k: K-means得到的anchor k
        :return k: input
        """
        # 将K-means得到的anchor k按面积从小到大排序
        k = k[np.argsort(k.prod(1))]
        # x: [N, 9] N个 ground truth 框与所有anchor框的宽比或高比(两者之中较小者)
        # best: [N] N个 ground truth 框与所有anchor框中的最大宽比或高比(两者之中较小者)
        x, best = metric(k, wh0)
        # (best > thr).float(): True=>1.  False->0.  .mean(): 求均值
        # BPR(best possible recall): 最多能被召回(通过thr)的 ground truth 框数量 / 所有 ground truth 框数量  [1] 0.96223  小于0.98 才会用K-means计算anchor
        # aat(anchors above threshold): [1] 3.54360 每个target平均有多少个anchors
        BPR, aat = (best > thr).float().mean(), (x > thr).float().mean() * n  # best possible recall, anch > thr
        f = anchor_fitness(k)
        # print(f'{prefix}thr={thr:.2f}: {BPR:.4f} best possible recall, {aat:.2f} anchors past thr')
        # print(f'{prefix}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, '
        #       f'past_thr={x[x > thr].mean():.3f}-mean: ', end='')
        print(f"aat: {aat:.5f}, fitness: {f:.5f}, best possible recall: {BPR:.5f}")
        for i, x in enumerate(k):
            print('%i,%i' % (round(x[0]), round(x[1])), end=',  ' if i = 2.0).any(1)]  # filter > 2 pixels
    # wh = wh * (np.random.rand(wh.shape[0], 1) * 0.9 + 0.1)  # multiply by random scale 0-1

    # Kmeans聚类方法: 使用欧式距离来进行聚类
    print(f'{prefix}Running kmeans for {n} anchors on {len(wh)} gt boxes...')
    # 计算宽和高的标准差->[w_std,h_std]
    s = wh.std(0)  # sigmas for whitening
    # 开始聚类,仍然是聚成 n 类,返回聚类后的anchors k(这个anchors k是白化后数据的anchor框s)
    # 另外还要注意的是这里的kmeans使用欧式距离来计算的
    # 运行K-means的次数为30次  obs: 传入的数据必须先白化处理 'whiten operation'
    # 白化处理: 新数据的标准差=1 降低数据之间的相关度，不同数据所蕴含的信息之间的重复性就会降低，网络的训练效率就会提高
    # 白化操作参考博客: https://blog.csdn.net/weixin_37872766/article/details/102957235
    k, dist = kmeans(wh / s, n, iter=30)  # points, mean distance
    assert len(k) == n, print(f'{prefix}ERROR: scipy.cluster.vq.kmeans requested {n} points but returned only {len(k)}')
    k *= s  # k*s 得到原来数据(白化前)的 anchor 框

    wh = flow.tensor(wh, dtype=flow.float32)  # filtered wh
    wh0 = flow.tensor(wh0, dtype=flow.float32)  # unfiltered wh0

    # 输出新算的anchors k 相关的信息
    k = print_results(k)

    # Plot wh
    # k, d = [None] * 20, [None] * 20
    # for i in tqdm(range(1, 21)):
    #     k[i-1], d[i-1] = kmeans(wh / s, i)  # points, mean distance
    # fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True)
    # ax = ax.ravel()
    # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
    # fig, ax = plt.subplots(1, 2, figsize=(14, 7))  # plot wh
    # ax[0].hist(wh[wh[:, 0]最佳适应度k 就进行选择操作
        if fg > f:
            # 选择变异后的anchor kg为最佳的anchor k 变异后的适应度fg为最佳适应度f
            f, k = fg, kg.copy()

            # 打印信息
            pbar.desc = f'{prefix}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
            if verbose:
                print_results(k)
    return print_results(k)

def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
    # Resize and pad image while meeting stride-multiple constraints
    shape = im.shape[:2]  # current shape [height, width] 当前图像尺寸
    if isinstance(new_shape, int):
        new_shape = (new_shape, new_shape)  # 缩放后的尺寸

    # Scale ratio (new / old) 计算缩放比例（选择长宽中更小的那个缩放比例）
    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
    if not scaleup:  # only scale down, do not scale up (for better val mAP)
        r = min(r, 1.0)

    # Compute padding
    ratio = r, r  # width, height ratios
    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))  # 直接缩放后的宽高
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding 计算灰边填充数值
    if auto:  # minimum rectangle 采用自适应图片缩放，确保宽和高都能被stride整除，因此需要补边
        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding 取余np.mod
    elif scaleFill:  # stretch  不采用自适应缩放，直接resize到目标shape，无需补边
        dw, dh = 0.0, 0.0
        new_unpad = (new_shape[1], new_shape[0])
        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios

    dw /= 2  # divide padding into 2 sides
    dh /= 2  # 上下和左右两侧各 padding 一半

    if shape[::-1] != new_unpad:  # resize
        im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))  # 上下两侧需要padding的大小
    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))  # 左右两侧需要padding的大小
    im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border 填充114
    return im, ratio, (dw, dh)

class Bottleneck(nn.Module):
    # Standard bottleneck
    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_, c2, 3, 1, g=g)
        self.add = shortcut and c1 == c2

    def forward(self, x):
        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))

class C3(nn.Module):
    # CSP Bottleneck with 3 convolutions
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)   # Conv = conv+BN+SiLU
        self.cv2 = Conv(c1, c_, 1, 1)
        self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)
        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))  # 串联n个残差结构
        # self.m = nn.Sequential(*(CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)))

    def forward(self, x):
        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))

class SPP(nn.Module):
    # Spatial Pyramid Pooling (SPP) layer https://arxiv.org/abs/1406.4729
    def __init__(self, c1, c2, k=(5, 9, 13)):  # 5, 9, 13为初始化的kernel size
        super().__init__()
        c_ = c1 // 2  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)          # 通道减半
        self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)  # concat之后的CBS
        self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])

    def forward(self, x):
        x = self.cv1(x)
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')  # suppress torch 1.9.0 max_pool2d() warning
            return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))


class SPPF(nn.Module):
    # Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
    def __init__(self, c1, c2, k=5):  # equivalent to SPP(k=(5, 9, 13))
        super().__init__()
        c_ = c1 // 2  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_ * 4, c2, 1, 1)
        self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)

    def forward(self, x):
        x = self.cv1(x)
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')  # suppress torch 1.9.0 max_pool2d() warning
            y1 = self.m(x)
            y2 = self.m(y1) # 串联k=5的池化，会获得9和13的池化，所以是等效的，但是时间更快
            return self.cv2(torch.cat((x, y1, y2, self.m(y2)), 1))

    def build_targets(self, p, targets):
        # Build targets for compute_loss(), input targets(image,class,x,y,w,h)
        na, nt = self.na, targets.shape[0]  # na为类别数，nt为目标数
        tcls, tbox, indices, anch = [], [], [], []
        gain = torch.ones(7, device=self.device)  # normalized to gridspace gain
        ai = torch.arange(na, device=self.device).float().view(na, 1).repeat(1, nt)  # ai.shape = (na, nt)，锚框的索引，第二个维度复制nt遍
        targets = torch.cat((targets.repeat(na, 1, 1), ai[..., None]), 2)  # targets.shape = (na, nt, 7)给每个目标加上锚框索引

        g = 0.5  # bias
        off = torch.tensor(
            [
                [0, 0],
                [1, 0],
                [0, 1],
                [-1, 0],
                [0, -1],  # j,k,l,m
                # [1, 1], [1, -1], [-1, 1], [-1, -1],  # jk,jm,lk,lm
            ],
            device=self.device).float() * g  # offsets

        for i in range(self.nl):  # self.nl为预测层也就是检测头的数量，anchor匹配需要逐层进行
            anchors = self.anchors[i]  # 该预测层上的anchor尺寸，三个尺寸
            gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]]  # 比如在P3层 gain=tensor([ 1.,  1., 80., 80., 80., 80.,  1.], device='cuda:0')

            # Match targets to anchors
            t = targets * gain  # shape(3,n,7) 将归一化的gtbox乘以特征图尺度，将box坐标投影到特征图上
            if nt:
                # Matches
                r = t[..., 4:6] / anchors[:, None]  # 计算标签box和当前层的anchors的宽高比，即:wb/wa,hb/ha
                j = torch.max(r, 1 / r).max(2)[0]  model.hyp['iou_t']  # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
                t = t[j]  # 筛选出符合条件target

                # Offsets
                gxy = t[:, 2:4]  # 得到相对于以左上角为坐标原点的坐标                          假设某个gt的中心点为gxy=[22.20, 19.05]
                gxi = gain[[2, 3]] - gxy  # 得到相对于右下角为坐标原点的坐标                   此时gxi=[17.80, 20.95]
                j, k = ((gxy % 1  1)).T  # jk判断gxy的中心点是否更偏向左上角     g=0.5 操作%1得到小数部分，小于0.5，所以j，k均为True
                l, m = ((gxi % 1  1)).T  # lm判断gxy的中心点是否更偏向右下角     g=0.5 l,m均为False，该舞台中心更偏向于左上角
                j = torch.stack((torch.ones_like(j), j, k, l, m))  # 网格本身是True，再加上 上下左右
                t = t.repeat((5, 1, 1))[j]  # 这里将t复制5个，然后使用j来过滤
                offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
            else:
                t = targets[0]
                offsets = 0

            # Define
            bc, gxy, gwh, a = t.chunk(4, 1)  # (image, class), grid xy, grid wh, anchors
            a, (b, c) = a.long().view(-1), bc.long().T  # anchors, image, class 其中，a表示当前gt box和当前层的第几个anchor匹配上了
            gij = (gxy - offsets).long()  # .long()为取整 gij是gxy的整数部分
            gi, gj = gij.T  # grid indices (gi,gj)是我们计算出来的负责预测该gt box的网格的坐标。

            # Append
            # indices中是正样本所对应的gt的信息  b表示当前正样本对应的gt属于该batch内第几张图片，a表示gtbox与anchors的对应关系，gj负责预测的网格纵坐标，gi负责预测的网格横坐标
            indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1)))  # image, anchor, grid indices
            # tbox, anch, tcls是正样本自己的信息
            tbox.append(torch.cat((gxy - gij, gwh), 1))  # 正样本相对网格的偏移，宽高
            anch.append(anchors[a])  # 正样本对应的anchor信息
            tcls.append(c)  # 正样本的类别信息

        return tcls, tbox, indices, anch

# Classification
if self.nc > 1:  # cls loss (only if multiple classes)
    t = torch.full_like(pcls, self.cn, device=self.device)  # torch.full_like返回一个形状与pcls相同且值全为self.cn的张量
    t[range(n), tcls[i]] = self.cp  # 对应类别处为self.cp， 其余类别处为self.cn
    lcls += self.BCEcls(pcls, t)  # BCE

                # pxy, pwh, _, pcls = pi[b, a, gj, gi].tensor_split((2, 4, 5), dim=1)  # faster, requires torch 1.8.0
                pxy, pwh, _, pcls = pi[b, a, gj, gi].split((2, 2, 1, self.nc), 1)  # target-subset of predictions

                # Regression
                pxy = pxy.sigmoid() * 2 - 0.5
                pwh = (pwh.sigmoid() * 2) ** 2 * anchors[i]
                pbox = torch.cat((pxy, pwh), 1)  # predicted box
                iou = bbox_iou(pbox, tbox[i], CIoU=True).squeeze()  # iou(prediction, target)
                lbox += (1.0 - iou).mean()  # iou loss

                # Objectness
                iou = iou.detach().clamp(0).type(tobj.dtype)
                if self.sort_obj_iou:
                    j = iou.argsort()
                    b, a, gj, gi, iou = b[j], a[j], gj[j], gi[j], iou[j]
                if self.gr 

def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
    # Returns Intersection over Union (IoU) of box1(1,4) to box2(n,4)

    # Get the coordinates of bounding boxes
    if xywh:  # transform from xywh to xyxy
        (x1, y1, w1, h1), (x2, y2, w2, h2) = box1.chunk(4, 1), box2.chunk(4, 1)
        w1_, h1_, w2_, h2_ = w1 / 2, h1 / 2, w2 / 2, h2 / 2
        b1_x1, b1_x2, b1_y1, b1_y2 = x1 - w1_, x1 + w1_, y1 - h1_, y1 + h1_
        b2_x1, b2_x2, b2_y1, b2_y2 = x2 - w2_, x2 + w2_, y2 - h2_, y2 + h2_
    else:  # x1, y1, x2, y2 = box1
        b1_x1, b1_y1, b1_x2, b1_y2 = box1.chunk(4, 1)
        b2_x1, b2_y1, b2_x2, b2_y2 = box2.chunk(4, 1)
        w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
        w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps

    # Intersection area
    inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * 
            (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)

    # Union Area
    union = w1 * h1 + w2 * h2 - inter + eps

    # IoU
    iou = inter / union
    if CIoU or DIoU or GIoU:
        cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1)  # convex (smallest enclosing box) width
        ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1)  # convex height
        if CIoU or DIoU:  # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
            c2 = cw ** 2 + ch ** 2 + eps  # convex diagonal squared
            rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 + (b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4  # center dist ** 2
            if CIoU:  # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
                v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
                with torch.no_grad():
                    alpha = v / (v - iou + (1 + eps))
                return iou - (rho2 / c2 + v * alpha)  # CIoU
            return iou - rho2 / c2  # DIoU
        c_area = cw * ch + eps  # convex area
        return iou - (c_area - union) / c_area  # GIoU https://arxiv.org/pdf/1902.09630.pdf
    return iou  # IoU