图片数据增强

from keras.preprocessing import image
import imgaug as ia
from imgaug import augmenters as iaa


sometimes = lambda aug: iaa.Sometimes(0.9, aug)
seq = iaa.Sequential([
    iaa.Fliplr(0.5),
    sometimes(
        iaa.OneOf([
            iaa.Affine(
                rotate=(-10, 10), # rotate by -45 to +45 degrees
                cval=0 # if mode is constant, use a cval between 0 and 255
            ),
            iaa.AddToHueAndSaturation((-20, 20)),
            iaa.Add((-20, 20), per_channel=0.5),
            iaa.Multiply((0.8, 1.2), per_channel=0.5),
            iaa.GaussianBlur((0, 0.5)), # blur images with a sigma between 0 and 3.0
            iaa.ContrastNormalization((0.5, 1.5), per_channel=0.5), # improve or worsen the contrast
            iaa.Sharpen(alpha=(0, 0.3), lightness=(0.7, 1.3)), # sharpen images
            iaa.Emboss(alpha=(0, 0.5), strength=(0, 0.5)) # emboss images
        ])
    ),
    iaa.Crop(percent=(0, 0.1))
],random_order=True)

def train_generator(df, batch_size):
    while True:
        df = df.sample(frac=1).reset_index(drop=True)
        for start in range(0, df.shape[0], batch_size):
            end = min(start + batch_size, df.shape[0])
            sub_df = df.iloc[start:end,:]
            x_batch = []
            y_batch = []
            for index, row in sub_df.iterrows():
                img_path = '../../data/images/train/%d.jpg'%row['imageId']
                img = cv2.imread(img_path)
                img = cv2.resize(img,(SIZE, SIZE), interpolation = cv2.INTER_CUBIC)
                img = seq.augment_image(img)
                img = image.img_to_array(img)
                img = preprocess_input(img)
                x_batch.append(img)
                y_batch.append(row[0:NUMBER_OF_CLASSES])
            yield np.array(x_batch), np.array(y_batch)