第一次比较完整地参加在线的训练营，也相对认真地实现了课程作业。通过这次训练营，比较直观地了解了图像分割网络的基本原理，并对几个经典的网络做了代码实现。
其中第一次作业的图像数据增强部分代码实现留下的印象比较深，涉及到一些图像处理和Numpy数组操作的技巧，自己纯手工实现，贴出来以供参考。
import cv2
import numpy as npclass Compose(object):def __init__(self, transforms):self.transforms = transformsdef __call__(self, image, label=None):for t in self.transforms:image, label = t(image, label)return image, labelclass Normalize(object):def __init__(self, mean_val, std_val, val_scale=1):# set val_scale = 1 if mean and std are in range (0,1)# set val_scale to other value, if mean and std are in range (0,255)self.mean = np.array(mean_val, dtype=np.float32)self.std = np.array(std_val, dtype=np.float32)self.val_scale = 1/255.0 if val_scale==1 else 1def __call__(self, image, label=None):image = image.astype(np.float32)image = image * self.val_scaleimage = image - self.meanimage = image * (1 / self.std)return image, labelclass ConvertDataType(object):def __call__(self, image, label=None):if label is not None:label = label.astype(np.int64)return image.astype(np.float32), labelclass Pad(object):def __init__(self, size, ignore_label=255, mean_val=0, val_scale=1):# set val_scale to 1 if mean_val is in range (0, 1)# set val_scale to 255 if mean_val is in range (0, 255) factor = 255 if val_scale == 1 else 1self.size = sizeself.ignore_label = ignore_labelself.mean_val=mean_val# from 0-1 to 0-255if isinstance(self.mean_val, (tuple,list)):self.mean_val = [int(x* factor) for x in self.mean_val]else:self.mean_val = int(self.mean_val * factor)def __call__(self, image, label=None):h, w, c = image.shapepad_h = max(self.size - h, 0)pad_w = max(self.size - w, 0)pad_h_half = int(pad_h / 2)pad_w_half = int(pad_w / 2)if pad_h > 0 or pad_w > 0:image = cv2.copyMakeBorder(image,top=pad_h_half,left=pad_w_half,bottom=pad_h - pad_h_half,right=pad_w - pad_w_half,borderType=cv2.BORDER_CONSTANT,value=self.mean_val)if label is not None:label = cv2.copyMakeBorder(label,top=pad_h_half,left=pad_w_half,bottom=pad_h - pad_h_half,right=pad_w - pad_w_half,borderType=cv2.BORDER_CONSTANT,value=self.ignore_label)return image, label# TODO
class CenterCrop(object):def __init__(self, size):self.size = sizedef __call__(self, image, label):h, w = image.shape[:2]h_start = int((h - self.size) / 2)w_start = int((w - self.size) / 2)image = image[h_start:h_start+self.size, w_start:w_start+self.size, :]label = label[h_start:h_start+self.size, w_start:w_start+self.size, :]return image, label# TODO
class Resize(object):@staticmethoddef resize(srcimg, height, width):if(len(srcimg.shape) > 2):dstimg = cv2.resize(srcimg, (height, width), interpolation=cv2.INTER_NEAREST)else:dstimg = cv2.resize(srcimg, (height, width), interpolation=cv2.INTER_LINEAR)return dstimg# TODO
class RandomFlip(object):def __call__(self, image, label):if(np.random.randint(2) % 2 == 1):image = np.fliplr(image)label = np.fliplr(label)return image, label# TODO
class RandomCrop(object):def __init__(self, size=256):self.size = sizedef __call__(self, image, label):h, w = label.shape[:2]crop_y = np.random.randint(0, h-self.size)crop_x = np.random.randint(0, w-self.size)image = image[crop_y:crop_y + self.size, crop_x:crop_x + self.size, :]label = label[crop_y:crop_y + self.size, crop_x:crop_x + self.size]return image, label# TODO
class Scale(object):def __init__(self, scale):self.scale = scaledef __call__(self, image, label):h, w = image.shape[:2]h_new, w_new = int(h * self.scale), int(w * self.scale)image = Resize.resize(image, h_new, w_new)label = Resize.resize(label, h_new, w_new)centercrop = CenterCrop(h)return centercrop(image, label)# TODO
class RandomScale(object):def __init__(self, scale_max):self.scale_max = scale_maxdef __call__(self, image, label):scale = 1.0 + np.random.rand() * (self.scale_max - 1.0)h, w = image.shape[:2]h_new, w_new = int(h * scale), int(w * scale)image = Resize.resize(image, h_new, w_new)label = Resize.resize(label, h_new, w_new)centercrop = CenterCrop(h)return centercrop(image, label)def main():image = cv2.imread('./work/dummy_data/JPEGImages/2008_000064.jpg')label = cv2.imread('./work/dummy_data/GroundTruth_trainval_png/2008_000064.png')result_dir = './work/transformed/'# TODO: crop_sizeimage = Resize.resize(image, 480, 640)label = Resize.resize(label, 480, 640)# TODO: Transform: RandomSacle, RandomFlip, Pad, RandomCropcompose = Compose([RandomCrop(256),RandomScale(2.0),RandomFlip(),Pad(256)])for i in range(10):# TODO: call transformdstimg, dstlab = compose(image, label)# TODO: save imagecv2.imwrite(result_dir + str(i) + '_image.jpg', dstimg)cv2.imwrite(result_dir + str(i) + '_label.jpg', dstlab)if __name__ == "__main__":main()课程链接：https://aistudio.baidu.com/aistudio/education/group/info/1767