刚刚使用tensorflow自带的dataset类做数据读取,非常的方便,功能也很强大。终于不需要用队列了。
def _parse_function(filename, label):
image_string = tf.read_file(filename)
image_decoded = tf.image.decode_image(image_string)
#image_resized = tf.image.resize_images(image_decoded, [28, 28])
return image_decoded, label
chars, char_dict, char_dict_re, train_f_name_list, test_f_name_list, len_test_f, len_train_f = data_init()
def dataset_test():
# 获取图片名
tra = tf.constant([imagepath+ '/'+test_f_name_list[i].split()[0] for i in range(100)])
# 获取label
labels = tf.constant([test_f_name_list[i].split()[1:] for i in range(100)])
# 将data和label给 dataset 对象
tra_data = tf.data.Dataset.from_tensor_slices((tra, labels))
# 读取data字符串指定的磁盘上的数据,通过_parse_function
tra_data = tra_data.map(_parse_function)
# 设置batch size
tra_data = tra_data.batch(3)
# 申请iterator
iterator = tra_data.make_initializable_iterator()
# 得到数据
next_element = iterator.get_next()
with tf.Session() as sess:
sess.run(iterator.initializer)
t0 = time.time()
for i in range(20):
data = sess.run(next_element)
tt = time.time()
print('time:%f shape:%s' % (tt-t0, data[0].shape))
t0 = tt
dataset_test()
'''
读取速度挺快的
time:0.002507 shape:(3, 32, 280, 3)
time:0.002005 shape:(3, 32, 280, 3)
time:0.002005 shape:(3, 32, 280, 3)
'''
参考文章:URL: https://blog.csdn.net/west_609/article/details/78608541