【yolov5 6.0 源码解析】---utils /augmentations.py

yolov5 数据增强代码

主要有以下几种方式：

class Albumentations # 数据增强package,比pytorch 自带的transform 更丰富
def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5) # 图像增强方式，hgain 是色调，不同色调不同颜色，sgain是饱和度， vgain是亮度
def hist_equalize(im, clahe=True, bgr=False):# 采用自适应直方图均衡化做图像增强
def replicate(im, labels) # 
def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32) # 图像size扩充至指定大小
def random_perspective(im, targets=(), segments=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0,
                       border=(0, 0)) # 随机增强
def copy_paste(im, labels, segments, p=0.5)# 复制粘贴
def cutout(im, labels, p=0.5) # 裁剪
def mixup(im, labels, im2, labels2) # mixup 
def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1, eps=1e-16) # 框筛选

下面一个个来看图像增强的方式：

Albumentations 图像增强

class Albumentations:
    # YOLOv5 Albumentations class (optional, only used if package is installed)
    def __init__(self):
        self.transform = None
        try:
            '''
            albumentations --一个数据增强的package,比pytorch的transform丰富；详情
            https://blog.csdn.net/cp1314971/article/details/106039800?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522164015856916780261966386%2522%252C%2522scm%2522%253A%252220140713.130102334..%2522%257D&request_id=164015856916780261966386&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~baidu_landing_v2~default-1-106039800.pc_search_es_clickV2&utm_term=import+albumentations+&spm=1018.2226.3001.4187
            '''
            import albumentations as A
            check_version(A.__version__, '1.0.3')  # version requirement 

            self.transform = A.Compose([
                A.Blur(p=0.01), # 图像随机大小内核模糊输入图像
                A.MedianBlur(p=0.01), # 图像随机模糊输入图像
                A.ToGray(p=0.01), # 转成灰度图 
                A.CLAHE(p=0.01), # 
                A.RandomBrightnessContrast(p=0.0), # 随机亮度和对比度
                A.RandomGamma(p=0.0), # 
                A.ImageCompression(quality_lower=75, p=0.0)], # 图像压缩 
                bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels'])) # 

            logging.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:
            logging.info(colorstr('albumentations: ') + f'{e}')

    def __call__(self, im, labels, p=1.0):
        if self.transform and random.random() < p:
            new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0])  # transformed
            im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
        return im, labels

hsv 色调-饱和度-亮度的图像增强

def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5): # 做h-色调， s-饱和度， v-亮度上面的随机增强 
    # HSV color-space augmentation
    if hgain or sgain or vgain:
        r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1  # random gains 生成3个[-1, 1)之间的随机数，分别与hsv相乘后+1 [0,2]之间
        hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV)) # 将图像从BGR 转成HSV ，拆分
        dtype = im.dtype  # uint8 

        x = np.arange(0, 256, dtype=r.dtype) # [0, 1, 2,...,255]
        lut_hue = ((x * r[0]) % 180).astype(dtype) #  [0, 180) 
        lut_sat = np.clip(x * r[1], 0, 255).astype(dtype) # 将数组截断至[0, 255]
        lut_val = np.clip(x * r[2], 0, 255).astype(dtype) # 将数组截断至[0, 255] 

        im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val))) 
        # cv2.LUT lookup-table 查找表方式，即通过lut_hue 这个表对之前hue数值做修正，返回0-255对应位置的lut_hue值  具体： https://blog.csdn.net/Dontla/article/details/103963085
        # cv2.merge 合并三个通道 
        cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im)  # no return needed

直方图均衡化增强

def hist_equalize(im, clahe=True, bgr=False): # 直方图均衡化增强 参考 https://www.cnblogs.com/my-love-is-python/p/10405811.html
    # Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
    yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV) # 将图像从bgr转成YUV 
    if clahe:
        c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8)) 
        # cv2.createCLAHE 实例化自适应直方图均衡化函数 局部直方图均衡化 ，不会使得细节消失 
        # c.apply 进行自适应直方图均衡化 
        yuv[:, :, 0] = c.apply(yuv[:, :, 0])
    else:
        # cv2.equalizeHist 进行像素点的均衡化 ，即全局均衡化 ，使得整体亮度提升，但是局部会模糊 
        yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0])  # equalize Y channel histogram
    return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB)  # convert YUV image to RGB

图像框的平移复制增强

def replicate(im, labels): # 复制，实际上指的是框的平移 
    # Replicate labels 
    h, w = im.shape[:2] # 获取图像长宽 
    boxes = labels[:, 1:].astype(int) # 获取框的位置和大小 
    x1, y1, x2, y2 = boxes.T # 框的左右和上下位置 
    s = ((x2 - x1) + (y2 - y1)) / 2  # side length (pixels)
    for i in s.argsort()[:round(s.size * 0.5)]:  # smallest indices
        x1b, y1b, x2b, y2b = boxes[i]
        bh, bw = y2b - y1b, x2b - x1b 
        yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw))  # offset x, y
        x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
        im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b]  # im4[ymin:ymax, xmin:xmax]
        labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)

    return im, labels

图像以letterbox缩放

def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
    # 按比例缩放图片，并将其他部分填充，到resize图片的大小 
    # Resize and pad image while meeting stride-multiple constraints
    shape = im.shape[:2]  # current shape [height, width]
    if isinstance(new_shape, int): # 如果输入是一个数字，默认长宽相等
        new_shape = (new_shape, new_shape)

    # Scale ratio (new / old)
    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) # 如果只缩小，不放大图片
        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)) # 对图片按比例缩放后的长宽 (width, height)
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding 对缩放后的图像 需要填充的size 
    if auto:  # minimum rectangle
        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding 取能被stride 整除的dw 和dh 
    elif scaleFill:  # stretch
        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

    if shape[::-1] != new_unpad:  # resize
        im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR) # 先将图片按比例缩放到指定大小
    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1)) # 上下位置
    left, right = int(round(dw - 0.1)), int(round(dw + 0.1)) # 左右位置 
    im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
    # cv2.copyMakeBorder 对im设置边界框
    return im, ratio, (dw, dh)

旋转等变换（未更新完全，后续补充）

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=(.1, .1), scale=(.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 [w, h, c] -->[w/2 , h/2, c]
    '''
    [ 1  0 -w/2
      0  1 -h/2
      0  0   1
    ]
    '''
    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 [w, h, c] -->[w/2 , h/2, c]
    '''
    [   1   0  0
      rand  1  0
      rand  0  1
    ]
    '''
    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 tf.matmul(A,C)=np.dot(A,C)= A@C 
    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))
        else:  # affine
            im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114)) # cv2.warpAffine()放射变换函数，可实现旋转，平移，缩放；变换后的平行线依旧平行 
    # cv2.warpAffine()放射变换函数，可实现旋转，平移，缩放；变换后的平行线依旧平行 
    # cv2.warpPerspective()透视变换函数，可保持直线不变形，但是平行线可能不再平行 

    # 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
        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

复制粘贴

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 < 0.30).all():  # allow 30% obscuration of existing labels
                labels = np.concatenate((labels, [[l[0], *box]]), 0)
                segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
                cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED)

        result = cv2.bitwise_and(src1=im, src2=im_new)
        result = cv2.flip(result, 1)  # augment segments (flip left-right)
        i = result > 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

图片随机马赛克

def cutout(im, labels, p=0.5):# 随机马赛克 
    # Applies image cutout augmentation https://arxiv.org/abs/1708.04552
    if random.random() < p:
        h, w = im.shape[:2]
        scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16  
        # [0.5 0.25 0.25 0.125 0.125 0.0125 ... 0.03125]  31
        # image size fraction
        for s in scales:
            mask_h = random.randint(1, int(h * s))  # create random masks
            mask_w = random.randint(1, int(w * s))  

            # box 确定随机马赛克的位置 
            xmin = max(0, random.randint(0, w) - mask_w // 2)
            ymin = max(0, random.randint(0, h) - mask_h // 2)
            xmax = min(w, xmin + mask_w)
            ymax = min(h, ymin + mask_h)

            # apply random color mask 确定随机马赛克的马赛克像素
            im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)] # 

            # return unobscured labels
            if len(labels) and s > 0.03: # 
                box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
                ioa = bbox_ioa(box, labels[:, 1:5])  # intersection over area 计算马赛克位置与框位置的IOU 大小 
                labels = labels[ioa < 0.60]  # remove >60% obscured labels # 保留下与马赛克框iou小于 0.6的，大于0.6的认为没有学习的必须 

    return labels

mixup

def mixup(im, labels, im2, labels2): # 两张图片像素点以不同权重融合 
    # 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) # img 做融合 
    labels = np.concatenate((labels, labels2), 0)
    return im, labels

过滤图像

def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1, eps=1e-16):  # box1(4,n), box2(4,n)
    # 过滤掉长宽小于阈值，面积图片占比小于阈值，框长宽比大于阈值的框 
    # Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
    # box1 图片 ，box2 框
    w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
    w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
    ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps))  # aspect ratio 宽长比与长宽比的最大值 
    return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr)  # candidates