In my experiment, I use feed to fill in the data. The code at sess is as follows:

1 with tf.Session() as sess:
2     init = tf.global_variables_initializer()
3     sess.run(init)
4     for epoch in range(a.epochs):
5         input, target = load_batch_data(batch_size=16, a=a)
6         batch_input = input.astype(np.float32)
7         batch_target = target.astype(np.float32)
8         sess.run(predict_real, feed_dict={input: batch_input, target: batch_target})

When running: {typeerror} unhashable type: ‘numpy. Ndarray’

Later, we found that:

When defining input and target outside the session, it is written as follows:

1 input = tf.placeholder(dtype=tf.float32, shape=[None, image_size, image_size, num_channels])
2 target = tf.placeholder(dtype=tf.float32, shape=[None, image_size, image_size, num_channels])

However, I defined input, target after opening session. This results in me running the following line of code

1 sess.run(predict_real, feed_dict={input: batch_input, target: batch_target})

There is an error like {typeerror} unhashable type: ‘numpy. Ndarray’. However, this input and target are not input and target outside the session. If you know the reason, it’s easy to correct it. Just change the names of input and target in the session, as follows:

 1     with tf.Session() as sess:
 2         init = tf.global_variables_initializer()
 3         sess.run(init)
 4         if a.mode == 'train':
 5             for epoch in range(a.epochs):
 6                 batch_input, batch_target = load_batch_data(a=a)
 7                 batch_input = batch_input.astype(np.float32)
 8                 batch_target = batch_target.astype(np.float32)
 9                 sess.run(model, feed_dict={input: batch_input, target: batch_target})
10                 print('epoch' + str(epoch) + ':')
11             saver.save(sess, 'model_parameter/train.ckpt')
12             print('training finished!!!')
13         elif a.mode == 'test':
14             #ceshi
15             ckpt = tf.train.latest_checkpoint(a.checkpoint)
16             saver.restore(sess, ckpt)
17             # Get the image at the time of the test and add the label
18             batch_input, _ = load_batch_data(a=a)
19             # batch_input = batch_input/255.
20             batch_input = batch_input.astype(np.float32)
21             generator_output = sess.run(test_output, feed_dict={input: batch_input})
22             # The result is processed and 3 is subtracted from the image channel to obtain the rgb image
23             result = process_generator_output(generator_output)
24             if result:
25                 print('Done!')
26         else:
27             print('the MODE is not avaliable...')