有關於tensorflow的.TFRecords 檔案怎麼樣來生成和讀取操作
下面將介紹如何生成和讀取tfrecords檔案:
首先介紹tfrecords檔案的生成,直接上程式碼:
from random import shuffle import numpy as np import glob import tensorflow as tf import cv2 import sys import os # 因為我裝的是CPU版本的,執行起來會有'warning',解決方法入下,眼不見為淨~ os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' shuffle_data = True image_path = '/path/to/image/*.jpg' # 取得該路徑下所有圖片的路徑,type(addrs)= list addrs = glob.glob(image_path) # 標籤資料的獲得具體情況具體分析,type(labels)= list labels = ... # 這裡是打亂資料的順序 if shuffle_data: c = list(zip(addrs, labels)) shuffle(c) addrs, labels = zip(*c) # 按需分割資料集 train_addrs = addrs[0:int(0.7*len(addrs))] train_labels = labels[0:int(0.7*len(labels))] val_addrs = addrs[int(0.7*len(addrs)):int(0.9*len(addrs))] val_labels = labels[int(0.7*len(labels)):int(0.9*len(labels))] test_addrs = addrs[int(0.9*len(addrs)):] test_labels = labels[int(0.9*len(labels)):] # 上面不是獲得了image的地址麼,下面這個函式就是根據地址獲取圖片 def load_image(addr): # A function to Load image img = cv2.imread(addr) img = cv2.resize(img, (224, 224), interpolation=cv2.INTER_CUBIC) img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # 這裡/255是為了將畫素值歸一化到[0,1] img = img / 255. img = img.astype(np.float32) return img # 將資料轉化成對應的屬性 def _int64_feature(value): return tf.train.Feature(int64_list=tf.train.Int64List(value=[value])) def _bytes_feature(value): return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value])) def _float_feature(value): return tf.train.Feature(float_list=tf.train.FloatList(value=[value])) # 下面這段就開始把資料寫入TFRecods檔案 train_filename = '/path/to/train.tfrecords' # 輸出檔案地址 # 建立一個writer來寫 TFRecords 檔案 writer = tf.python_io.TFRecordWriter(train_filename) for i in range(len(train_addrs)): # 這是寫入操作視覺化處理 if not i % 1000: print('Train data: {}/{}'.format(i, len(train_addrs))) sys.stdout.flush() # 載入圖片 img = load_image(train_addrs[i]) label = train_labels[i] # 建立一個屬性(feature) feature = {'train/label': _int64_feature(label), 'train/image': _bytes_feature(tf.compat.as_bytes(img.tostring()))} # 建立一個 example protocol buffer example = tf.train.Example(features=tf.train.Features(feature=feature)) # 將上面的example protocol buffer寫入檔案 writer.write(example.SerializeToString()) writer.close() sys.stdout.flush()
接下來介紹tfrecords檔案的讀取:
import tensorflow as tf import numpy as np import matplotlib.pyplot as plt import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' data_path = 'train.tfrecords' # tfrecords 檔案的地址 with tf.Session() as sess: # 先定義feature,這裡要和之前建立的時候保持一致 feature = { 'train/image': tf.FixedLenFeature([], tf.string), 'train/label': tf.FixedLenFeature([], tf.int64) } # 建立一個佇列來維護輸入檔案列表 filename_queue = tf.train.string_input_producer([data_path], num_epochs=1) # 定義一個 reader ,讀取下一個 record reader = tf.TFRecordReader() _, serialized_example = reader.read(filename_queue) # 解析讀入的一個record features = tf.parse_single_example(serialized_example, features=feature) # 將字串解析成影象對應的畫素組 image = tf.decode_raw(features['train/image'], tf.float32) # 將標籤轉化成int32 label = tf.cast(features['train/label'], tf.int32) # 這裡將圖片還原成原來的維度 image = tf.reshape(image, [224, 224, 3]) # 你還可以進行其他一些預處理.... # 這裡是建立順序隨機 batches(函式不懂的自行百度) images, labels = tf.train.shuffle_batch([image, label], batch_size=10, capacity=30, min_after_dequeue=10) # 初始化 init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer()) sess.run(init_op) # 啟動多執行緒處理輸入資料 coord = tf.train.Coordinator() threads = tf.train.start_queue_runners(coord=coord) .... #關閉執行緒 coord.request_stop() coord.join(threads) sess.close()