用自己的數據集訓練Tensorflow模型
上篇博文我們用tensorflow實現了一些簡單的圖像處理
TensorFlow中的圖像處理
今天我們進一步來學習tensorflow
本文具體數據集與源代碼可從我的GitHub地址獲取
https://github.com/liuzuoping/Deep_Learning_note
- 數據預處理
- 數據的讀取
數據讀取
根據tensorflow的官方教程來看,tensorflow主要支持4中數據讀取的方式。
- Preloaded data: 預加載數據
- Feeding: 先產生一個個batch數據,然後在運行的時候依次餵給計算圖。
- Reading from file: 從文件中讀取,tensorflow主要支持兩種方法從文件中讀取數據餵給計算圖:一個是CSV,還有一個是TFRecords
- 多管線輸入
預加載數據
import tensorflow as tf
# 構建一個Graph
x1 = tf.constant([1, 2, 3])
x2 = tf.constant([4, 5, 6])
y = tf.add(x1, x2)
# 喂數據 -> 啓動session,計算圖
with tf.Session() as sess:
print(sess.run(y))
[5 7 9]
Feeding
import tensorflow as tf
# 構建Graph
x1 = tf.placeholder(tf.int16)
x2 = tf.placeholder(tf.int16)
y = tf.add(x1, x2)
# X_1,X_2是變量,可以賦予不同的值
X_1 = [1, 2, 3]
X_2 = [4, 5, 6]
# 喂數據 -> 啓動session,計算圖
with tf.Session() as sess:
print(sess.run(y, feed_dict={x1: X_1, x2: X_2}))
[5 7 9]
從文件中讀取
- 直接讀取文件
- 寫入TFRecord並讀取
直接讀取文件
# 導入tensorflow
import tensorflow as tf
# 新建一個Session
with tf.Session() as sess:
# 我們要讀三幅圖片plate1.jpg, plate2.jpg, plate3.jpg
filename = ['images/plate1.jpg', 'images/plate2.jpg', 'images/plate3.jpg']
# string_input_producer會產生一個文件名隊列
filename_queue = tf.train.string_input_producer(filename, shuffle=True, num_epochs=5)
# reader從文件名隊列中讀數據。對應的方法是reader.read
reader = tf.WholeFileReader()
key, value = reader.read(filename_queue)
# tf.train.string_input_producer定義了一個epoch變量,要對它進行初始化
tf.local_variables_initializer().run()
# 使用start_queue_runners之後,纔會開始填充隊列
threads = tf.train.start_queue_runners(sess=sess)
i = 0
while True:
i += 1
# 獲取圖片數據並保存
image_data = sess.run(value)
with open('tfIO/test_%d.jpg' % i, 'wb') as f:
f.write(image_data)
寫入TFRecord並讀取
import tensorflow as tf
# 爲顯示圖片
from matplotlib import pyplot as plt
import matplotlib.image as mpimg
%pylab inline
# 爲數據操作
import pandas as pd
import numpy as np
import os
img=mpimg.imread('images/plate1.jpg')
tensors = np.array([img,img,img])
# show image
print('\n張量')
display(tensors)
plt.imshow(img)
import os
import tensorflow as tf
from PIL import Image
def create_record():
writer = tf.python_io.TFRecordWriter("tfIO/tfrecord/test.tfrecord")
for i in range(3):
# 創建字典
features={}
# 寫入張量,類型float,本身是三維張量,另一種方法是轉變成字符類型存儲,隨後再轉回原類型
features['tensor'] = tf.train.Feature(bytes_list=tf.train.BytesList(value=[tensors[i].tostring()]))
# 存儲形狀信息(806,806,3)
features['tensor_shape'] = tf.train.Feature(int64_list = tf.train.Int64List(value=tensors[i].shape))
# 將存有所有feature的字典送入tf.train.Features中
tf_features = tf.train.Features(feature= features)
# 再將其變成一個樣本example
tf_example = tf.train.Example(features = tf_features)
# 序列化該樣本
tf_serialized = tf_example.SerializeToString()
# 寫入一個序列化的樣本
writer.write(tf_serialized)
# 由於上面有循環3次,所以到此我們已經寫了3個樣本
# 關閉文件
writer.close()
def read_and_decode(filename):
filename_queue = tf.train.string_input_producer([filename])
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(serialized_example,
features={
'tensor': tf.FixedLenFeature([], tf.string),
'tensor_shape' : tf.FixedLenFeature([], tf.int64),
})
tensor = tf.decode_raw(features['tensor'], tf.uint8)
tensor = tf.reshape(tensor, [224, 224, 3])
tensor = tf.cast(tensor, tf.float32) * (1. / 255) - 0.5
tensor_shape = tf.cast(features['tensor_shape'], tf.int32)
return tensor,tensor_shape
if __name__ == '__main__':
img, label = read_and_decode("tfIO/tfrecord/test.tfrecord")
img_batch, label_batch = tf.train.shuffle_batch([img, label],
batch_size=5, capacity=2000,
min_after_dequeue=1000)
#初始化所有的op
init = tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init)
#啓動隊列
threads = tf.train.start_queue_runners(sess=sess)
for i in range(3):
val, l= sess.run([img_batch, label_batch])
#l = to_categorical(l, 12)
print(val.shape, l)