checkpoint_hook = tf.train.CheckpointSaverHook(checkpoint_dir=summary_path, save_steps=saving_step, scaffold=scaffold)
config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=True, gpu_options=(tf.GPUOptions(allow_growth=True)))
session_creator = tf.train.ChiefSessionCreator(scaffold=scaffold, config=config, checkpoint_dir=summary_path)
# Create a session for running operations in the Graph
with tf.train.MonitoredSession(session_creator=session_creator, hooks=[checkpoint_hook]) as sess:
print("🤖 Start testing...")
step = 0
avg_loss = 0
while not sess.should_stop():
step += 1
l, pred, color, gray = sess.run([loss, predict_rgb, color_image_rgb, gray_image], feed_dict={is_training: False})
# Print batch loss
print("📖 Testing iter %d, Minibatch Loss = %f" % (step, l))
avg_loss += l
# Save all testing image
for i in range(batch_size):
summary_image = concat_images(gray[i], pred[i])
summary_image = concat_images(summary_image, color[i])
plt.imsave("%s/images/%d_%d.png" % (testing_summary, step, i), summary_image)
if step >= len(file_paths) / batch_size:
break
print("🎉 Testing finished!")
print("👀 Total average loss: %f" % (avg_loss / len(file_paths)))
# 创建一个会话来运行计算图的操作
with tf.train.MonitoredSession(session_creator=session_creator,
hooks=[checkpoint_hook]) as sess:
print("🤖 Start testing...")
step = 0
avg_loss = 0
while not sess.should_stop():
step += 1
l, pred, color, gray = sess.run(
[loss, predict_rgb, color_image_rgb, gray_image],
feed_dict={is_training: False})
# Print batch loss
print("📖 Testing iter %d, Minibatch Loss = %f" % (step, l))
avg_loss += l
# 保存所有的测试图片
for i in range(len(color)):
summary_image = concat_images(gray[i], pred[i])
# summary_image = concat_images(summary_image, color[i])
plt.imsave("%s/images/%d_%d.png" % (testing_summary, step, i),
summary_image)
# plt.imsave("%s/images/gray/%d_%d.jpeg" % (testing_summary, step, i), gray[i])
if step >= len(file_paths) / batch_size:
break
print("🎉 Testing finished!")
print("👀 Total average loss: %f" % (avg_loss / len(file_paths)))
[
cost, predict_rgb, color_image_rgb, gray_image_rgb,
merged
],
feed_dict={
is_training: False,
uv: 3
})
print "Iter %d, Minibatch Loss = %f" % (
step, float(np.mean(loss)))
train_writer.add_summary(summary, step)
train_writer.flush()
# Save test image
if step % test_step == 0:
summary_image = concat_images(gray[0], pred[0])
summary_image = concat_images(summary_image, color[0])
plt.imsave(
"%s/images/%s.jpg" % (train_summary, str(step)),
summary_image)
# Save model
if step % save_step == 0 and step != 0:
save_path = saver.save(sess, "%s/model.ckpt" % model_path)
print "Model saved in file: %s" % save_path
# Stop training
if step == training_iters:
break
print "Training Finished!"
avg_error = 0
while not coord.should_stop():
step += 1
# Print batch loss
if step % display_step == 0:
loss, pred, pred_rgb, color_rgb, gray_rgb, summary = \
sess.run([cost, predict, predict_rgb, color_image_rgb, gray_image_rgb, merged], feed_dict={is_training: False, uv: 3})
print "Iter %d, Minibatch Loss = %f" % (
step, float(np.mean(loss)))
avg_error += float(np.mean(loss))
test_writer.add_summary(summary, step)
test_writer.flush()
# Save output image
summary_image = concat_images(gray_rgb[0], pred_rgb[0])
summary_image = concat_images(summary_image, color_rgb[0])
plt.imsave("%s/images/%s.jpg" % (test_summary, str(step)),
summary_image)
if step == len(file_paths):
break
print "Testing Finished!"
print "Average error: %f" % (avg_error / len(file_paths))
sys.stdout.flush()
except tf.errors.OUT_OF_RANGE as e:
# Handle exception
print "Done training -- epoch limit reached"
coord.request_stop(e)
