tensorboard 可视化 mnist 数据集

原文链接: tensorboard 可视化 mnist 数据集

上一篇: tensorboard 写入图片

下一篇: tensorboard 可视化 mnist 预测失败的图片

将mnist 图像数据可视化到tensorboard 中

按照数字分类显示，可对预测结果进行人眼复查。。。。。。。。。。。。

写入时指定输出全部

            tf.summary.image(
                'img_' + str(i),
                new_img,
                max_outputs=cnt,  # 最多默认只输出3个
            )

参考代码

import tensorflow as tf
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

imgs = mnist.test.images
labels = mnist.test.labels

labels = np.argmax(labels, 1)

with tf.Session() as sess:
    summary_writer = tf.summary.FileWriter('logs/', sess.graph)

    sess.run(tf.global_variables_initializer())

    for i in range(10):
        cnt = labels[labels == i].shape[0]
        print(cnt)
        new_img = imgs[np.argwhere(labels == i)].reshape([cnt, 28, 28, 1])
        print(new_img.shape)
        img_summary = sess.run(
            tf.summary.image(
                'img_' + str(i),
                new_img,
                max_outputs=cnt,  # 最多默认只输出3个
            )
        )
        summary_writer.add_summary(img_summary)
        print('---------------')

测试集数据分类情况

980
(980, 28, 28, 1)
---------------
1135
(1135, 28, 28, 1)
---------------
1032
(1032, 28, 28, 1)
---------------
1010
(1010, 28, 28, 1)
---------------
982
(982, 28, 28, 1)
---------------
892
(892, 28, 28, 1)
---------------
958
(958, 28, 28, 1)
---------------
1028
(1028, 28, 28, 1)
---------------
974
(974, 28, 28, 1)
---------------
1009
(1009, 28, 28, 1)
---------------

tensorboard.exe  --logdir=logs