深度学习入门数据集--1.Cifar10数据集

import numpy as np  
from scipy.misc import imsave  
import matplotlib.pyplot as plt
import pylab

filename = '/Users/wangsen/ai/13/models-master/data/cifar-10-batches-bin/test_batch.bin' 
label_mate = '/Users/wangsen/ai/13/models-master/data/cifar-10-batches-bin/batches.meta.txt'
labels_txt = open(label_mate,"r").read().strip().split("\n")
bytestream = open(filename, "rb")  
buf = bytestream.read(25 * (1 + 32 * 32 * 3))  
bytestream.close()  

data = np.frombuffer(buf, dtype=np.uint8)  
data = data.reshape(25, 1 + 32*32*3)  
labels_images = np.hsplit(data, [1])  
labels = labels_images[0].reshape(25)  
images = labels_images[1].reshape(25, 32, 32, 3)  
fig, axes1 = plt.subplots(5,5,figsize=(4,5))
# for itr,label in enumerate(labels):
#         print(itr,":",labels_txt[label])
i = 0
for j in range(5):
    for k in range(5):
        img = np.reshape(images[i],(3,32,32))
        img = img.transpose(1,2,0)
        axes1[j][k].set_axis_off()
        axes1[j][k].imshow(img)
        axes1[j][k].set_title(labels_txt[labels[i]])
        i=i+1
pylab.show()

Filling queue with 20000 CIFAR images before starting to train. This will take a few minutes.
2019-02-20 13:42:05.167927: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-02-20 13:42:09.260566: step 0, loss = 4.67 (304.9 examples/sec; 0.420 sec/batch)
2019-02-20 13:42:13.762996: step 10, loss = 4.63 (284.3 examples/sec; 0.450 sec/batch)
2019-02-20 13:42:18.095651: step 20, loss = 4.49 (295.4 examples/sec; 0.433 sec/batch)
2019-02-20 13:42:22.444906: step 30, loss = 4.50 (294.3 examples/sec; 0.435 sec/batch)
2019-02-20 13:42:27.136578: step 40, loss = 4.40 (272.8 examples/sec; 0.469 sec/batch)
2019-02-20 13:42:31.833072: step 50, loss = 4.32 (272.5 examples/sec; 0.470 sec/batch)

A binary to train CIFAR-10 using a single GPU.

Accuracy:
cifar10_train.py achieves ~86% accuracy after 100K steps (256 epochs of
data) as judged by cifar10_eval.py.

Speed: With batch_size 128.

System        | Step Time (sec/batch)  |     Accuracy
------------------------------------------------------------------
1 Tesla K20m  | 0.35-0.60              | ~86% at 60K steps  (5 hours)
1 Tesla K40m  | 0.25-0.35              | ~86% at 100K steps (4 hours)

2019-02-20 15:59:41.109588: precision @ 1 = 0.606

# -*- coding:utf-8 -*-
import tensorflow as tf
from tensorflow.python.ops.image_ops_impl import ResizeMethod
from prettytable import PrettyTable  
 
import cifar10
import numpy as np
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
 
FLAGS = tf.app.flags.FLAGS
# 设置存储模型训练结果的路径
tf.app.flags.DEFINE_string('checkpoint_dir', '/Users/wangsen/ai/13/models-master/tutorials/image/cifar10/cifar10_train',
             """Directory where to read model checkpoints.""")
tf.app.flags.DEFINE_string('class_dir', '//Users/wangsen/ai/13/models-master/data/cifar-10-batches-bin/',
                           """存储文件batches.meta.txt的目录""")
tf.app.flags.DEFINE_string('test_file', '/Users/wangsen/Desktop/1.jpeg', """测试用的图片""")

IMAGE_SIZE = 24
 
 
def evaluate_images(images):  # 执行验证
    logits = cifar10.inference(images)
    load_trained_model(logits=logits)
 
 
def load_trained_model(logits):
    with tf.Session() as sess:
        ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
        if ckpt and ckpt.model_checkpoint_path:
            # 从训练模型恢复数据
            saver = tf.train.Saver()
            saver.restore(sess, ckpt.model_checkpoint_path)
        else:
            print('No checkpoint file found')
            return

 
        # 从文件以字符串方式获取10个类标签，使用制表格分割
        cifar10_class = np.loadtxt(FLAGS.class_dir + "batches.meta.txt", str, delimiter='\t')
        # 预测最大的三个分类
        top_k_pred = tf.nn.top_k(logits, k=3)
        output = sess.run(top_k_pred)
        probability = np.array(output[0]).flatten()  # 取出概率值，将其展成一维数组
        index = np.array(output[1]).flatten()
        # 使用表格的方式显示
        tabel = PrettyTable(["index", "class", "probability"])
        tabel.align["index"] = "l"  
        tabel.padding_width = 1 
        for i in np.arange(index.size):
            tabel.add_row([index[i], cifar10_class[index[i]], probability[i]])
        print(tabel)
    lena = mpimg.imread(FLAGS.test_file) # 读取和代码处于同一目录下的 lena.png
    plt.imshow(lena) # 显示图片
    plt.axis('off') # 不显示坐标轴
    plt.show() 
 
def img_read(filename):
    if not tf.gfile.Exists(filename):
        tf.logging.fatal('File does not exists %s', filename)
    image_data = tf.image.convert_image_dtype(tf.image.decode_jpeg(tf.read_file(filename),
                                                                   channels=3), dtype=tf.float32)
    height = IMAGE_SIZE
    width = IMAGE_SIZE
    image = tf.image.resize_images(image_data, (height, width), method=ResizeMethod.BILINEAR)
    image = tf.expand_dims(image, -1)
    image = tf.reshape(image, (1, 24, 24, 3))
    return image
 
def main(argv=None):  # pylint: disable=unused-argument
    filename = FLAGS.test_file
    images = img_read(filename)
    evaluate_images(images)
 
if __name__ == '__main__':
    tf.app.run()