yolo 系列数据增强总结

数据增强

代码主要位于class LoadImagesAndLabels 中。具体实现在方法getitem()中。

主要的增强方式有以下几种：

1. mosaic 增强

在代码中并不是超参中设置了hyp['mosaic'] 大于0的属性值，就可以进行mosaic的，具体执行还的进一步的判断：

mosaic = self.mosaic and random.random() < hyp['mosaic']

具体实现如下：(4 mosaic 方法，输入的index是图片的索引)

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
        #random.uniform()指定最小值和最大值范围内生成浮点数
        yc, xc = (int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border)  # mosaic center x, y
        #表示随机从图片序列中选出3张图片
        indices = [index] + random.choices(self.indices, k=3)  # 3 additional image indices
        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
                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)
                x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
​
            img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
            padw = x1a - x1b
            padh = y1a - y1b
​
            # Labels
            labels, segments = self.labels[index].copy(), self.segments[index].copy()
            if labels.size:
                labels[:, 1:] = xywh