分批读取训练数据进行训练

训练集数据量过大受设备内存影响不能将全部数据直接放到网络中进行训练，需要分批读取训练数据。

train_x为训练集地址，train_y为训练集标签，val_X为验证集数据，val_y为验证集标签。

分批读取函数如下：

def dataset_split(images, labels, batch_size):
    while 1:
        i = 0
        n = math.ceil(len(images)/batch_size)
        print(n)
        for j in range(n):
            if j != n-1:
                x = images[i : i + batch_size]
                y = labels[i : i + batch_size]
                i = i + batch_size
                X = []
                for m in range(len(x)):
                    a = cv2.imread(x[m])
                    #print(type(a))
                    a = a.tolist()
                    #print(type(a))
                    X.append(a)
                X = np.array(X)
                yield X, y
                
            if j == n-1:
                x = images[len(images)-batch_size: ]
                y = labels[len(labels)-batch_size: ]
                X = []
                for m in range(len(x)):
                    a = cv2.imread(x[m])
                    a = a.tolist()
                    X.append(a)
                X = np.array(X)
                yield X, y


model.fit_generator(dataset_split(train_x, train_y, batch_size),
                                        validation_data=(val_X, val_y), steps_per_epoch= 
                                        (len(train_x) // batch_size), epochs=EPOCHS, 
                                        verbose=2, callbacks=[csv_logger, checkpointer]) 

 

本文地址：https://blog.csdn.net/qq_21466543/article/details/107150645