这是一个完整的可运行示例，它重现了该问题：
import tensorflow as tftruth_filenames_np = ['dir/%d.jpg' % j for j in range(66)]truth_filenames_tf = tf.convert_to_tensor(truth_filenames_np)# get the labelslabels = [f.rsplit("/", 1)[1] for f in truth_filenames_np]labels_tf = tf.convert_to_tensor(labels)# My list is also already shuffled, so I set shuffle=Falsetruth_image_name, truth_label = tf.train.slice_input_producer( [truth_filenames_tf, labels_tf], shuffle=False)# # Another key step, where I batch them together# truth_images_batch, truth_label_batch = tf.train.batch(# [truth_image_name, truth_label], batch_size=11)epochs = 7with tf.Session() as sess: sess.run(tf.global_variables_initializer()) coord = tf.train.Coordinator() threads = tf.train.start_queue_runners(coord=coord) for i in range(epochs): print("Epoch ", i) X_truth_batch = truth_image_name.eval() X_label_batch = truth_label.eval() # Here I display all the images in this batch, and then I check # which file numbers they actually are. # BUT, the images that are displayed don't correspond with what is # printed by X_label_batch! print(X_truth_batch) print(X_label_batch) coord.request_stop() coord.join(threads)打印的内容是：
Epoch 0b'dir/0.jpg'b'1.jpg'Epoch 1b'dir/2.jpg'b'3.jpg'Epoch 2b'dir/4.jpg'b'5.jpg'Epoch 3b'dir/6.jpg'b'7.jpg'Epoch 4b'dir/8.jpg'b'9.jpg'Epoch 5b'dir/10.jpg'b'11.jpg'Epoch 6b'dir/12.jpg'b'13.jpg'
因此，基本上每个eval调用都会再次运行该操作！添加批处理对此没有任何影响-只是打印批处理（前11个文件名，后11个标签，依此类推）
我看到的解决方法是：
for i in range(epochs): print("Epoch ", i) pair = tf.convert_to_tensor([truth_image_name, truth_label]).eval() print(pair[0]) print(pair[1])正确打印：
Epoch 0b'dir/0.jpg'b'0.jpg'Epoch 1b'dir/1.jpg'b'1.jpg'# ...
但对于违反最不惊奇原则的行为却无能为力。
编辑 ：另一种方法：
import tensorflow as tftruth_filenames_np = ['dir/%d.jpg' % j for j in range(66)]truth_filenames_tf = tf.convert_to_tensor(truth_filenames_np)labels = [f.rsplit("/", 1)[1] for f in truth_filenames_np]labels_tf = tf.convert_to_tensor(labels)truth_image_name, truth_label = tf.train.slice_input_producer( [truth_filenames_tf, labels_tf], shuffle=False)epochs = 7with tf.Session() as sess: sess.run(tf.global_variables_initializer()) tf.train.start_queue_runners(sess=sess) for i in range(epochs): print("Epoch ", i) X_truth_batch, X_label_batch = sess.run( [truth_image_name, truth_label]) print(X_truth_batch) print(X_label_batch)这是一个更好的方法，因为
tf.convert_to_tensor并且co只接受相同类型/形状等的张量。
请注意，为简单起见，我删除了协调器，但是会导致警告：
W c： tf_jenkins home workspace release-win device cpu os
windows tensorflow core kernels queue_base.cc：294] _0_input_producer
/ input_producer / fraction_of_32_full / fraction_of_32_full：跳过未关闭队列的取消入队尝试
看到这个
