keras从dataframe中读取数据并进行数据增强进行训练（分类+分割）

keras的数据读取并进行在线数据增强进行训练，此处使用pandas dataframe的形式～

dataframe 的数据存储

分类

迭代器生成数据

def make_image_gen(in_df, batch_size = BATCH_SIZE):
    file = in_df['filepath'].values
    label = in_df['label'].values
    all_batches = np.stack((file, label),1)
    out_rgb = []
    out_y = []
    while True:
        np.random.shuffle(all_batches)
        for data in all_batches:
            file = data[0]
            c_img = cv2.imread(file)
            out_rgb += [c_img]
            out_y.append([data[1]]) 
            if len(out_rgb)>=batch_size:
                yield np.stack(out_rgb, 0)/255.0, np.array(out_y)
                out_rgb, out_y = [], []
                

数据增强使用ImageDataGenerator

from keras.preprocessing.image import ImageDataGenerator
dg_args = dict(featurewise_center = False, 
                  samplewise_center = False,
                  rotation_range = 5, 
                  width_shift_range = 0.01, 
                  height_shift_range = 0.01, 
                  shear_range = 0.01,
                  zoom_range = [0.9, 1.25],  
                  horizontal_flip = True, 
                  vertical_flip = True, ###
                  fill_mode = "constant",
                   cval = 0,
                   data_format = 'channels_last')

image_gen = ImageDataGenerator(**dg_args)
def create_aug_gen(in_gen):
    for in_x, in_y in in_gen:
        a = image_gen.flow(255*in_x,in_y,batch_size = in_x.shape[0])
        x, y = next(a)
        yield x/255.0, y
        

训练

aug_gen = create_aug_gen(make_image_gen(data_train))
model.fit_generator(aug_gen, 
                    steps_per_epoch=1, 
                    epochs=1)

分割

迭代器生成数据

def make_image_gen(in_df, batch_size = BATCH_SIZE):
    file = in_df['filepath'].values
    mask = in_df['maskpath'].values
    all_batches = np.stack((file,mask),1)
    out_rgb = []
    out_mask = []
    while True:
        np.random.shuffle(all_batches)
        for data in all_batches:
            file = data[0]
            c_img = cv2.imread(file)
            mask = data[1]
            c_mask = cv2.imread(mask, 0)
            c_mask = np.expand_dims(c_mask, axis=-1)
            out_rgb += [c_img]
            out_mask += [c_mask]
            if len(out_rgb)>=batch_size:
                yield np.stack(out_rgb, 0)/255.0, np.stack(out_mask, 0)
                out_rgb, out_mask=[], []

图像和掩码进行相同的数据增强

from keras.preprocessing.image import ImageDataGenerator
dg_args = dict(featurewise_center = False, 
                  samplewise_center = False,
                  rotation_range = 45, 
                   data_format = 'channels_last')

image_gen = ImageDataGenerator(**dg_args)
label_gen = ImageDataGenerator(**dg_args)
def create_aug_gen(in_gen, seed = None):
    np.random.seed(seed if seed is not None else np.random.choice(range(9999)))
    for in_x, in_y in in_gen:
        seed = np.random.choice(range(9999))
        g_x = image_gen.flow(255*in_x, 
                             batch_size = in_x.shape[0], 
                             seed = seed, 
                             shuffle=True)
        g_y = label_gen.flow(in_y, 
                             batch_size = in_x.shape[0], 
                             seed = seed, 
                             shuffle=True)

        yield next(g_x)/255.0, next(g_y)

训练

valid_x, valid_y = next(make_image_gen(data_val_mb, 1))
aug_gen = create_aug_gen(make_image_gen(data_train_mb))
model.fit_generator(aug_gen, 
                    steps_per_epoch=1, 
                    epochs=NB_EPOCHS, 
                    validation_data=(valid_x, valid_y))