Tensorflow的可视化操作

如何通过浏览器查看结果

• 首先我们通过TF将我们的训练对应的图，进行点保存

  summary_writer = tf.summary.FileWriter('E:\TensorFlowMINST\\tensorboard', sess.graph)

数据结果如图所示：

• 开启Tensorflow的服务器，进行查看我们的图。

tensorboard  --logdir  #你保存图的目录

如何进行保存

1. 首先我们需要使用 tf.summary.下的api进行保存你想要存入的内容。
2. 在保存前，我们需要声明我们的标记tf.summary.merge_all()
3. 使用tf.summary.FileWriter('你想保存的路径',sess.grap)进行设置我们图所生成文件的路径。
4. 余下的就是讲内容写入文件中，

import tensorflow as tfPicPath = 'timg.jpg'reader = tf.WholeFileReader()key, value = reader.read(tf.train.string_input_producer([PicPath]))image0 = tf.image.decode_jpeg(value)image = tf.expand_dims(image0, 0)image_summary = tf.summary.image('orign image', image)histogram_summary = tf.summary.histogram('image hist', image)e = tf.reduce_mean(image)scalar_summary = tf.summary.scalar('image mean', e)resized_image = tf.image.resize_images(image, [256, 256], method=tf.image.ResizeMethod.AREA)img_resize_summary = tf.summary.image('image resized', resized_image)cropped_image = tf.image.crop_to_bounding_box(image0, 20, 20, 256, 256)cropped_image_summary = tf.summary.image('image cropped', tf.expand_dims(cropped_image, 0))flipped_image = tf.image.flip_left_right(image0)flipped_image_summary = tf.summary.image('image flipped', tf.expand_dims(flipped_image, 0))rotated_image = tf.image.rot90(image0, k=1)rotated_image_summary = tf.summary.image('image rotated', tf.expand_dims(rotated_image, 0))grayed_image = tf.image.rgb_to_grayscale(image0)grayed_image_summary = tf.summary.image('image grayed', tf.expand_dims(grayed_image, 0))merged = tf.summary.merge_all()init_op = tf.initialize_all_variables()with tf.Session() as sess:    print(sess.run(init_op))    cord = tf.train.Coordinator()    threads = tf.train.start_queue_runners(coord=cord)    img = image.eval();    print(img.shape)    cord.request_stop()    cord.join(threads)    # 声明需要写入图的文件路径    summary_writer = tf.summary.FileWriter('E:\TensorFlowMINST\\tensorboard', sess.graph)    # 获取所有的标注点    summary_all = sess.run(merged)    # 增加标志点进入图文件中    summary_writer.add_summary(summary_all, 0)    summary_writer.close()