tensorflowの停止条件が分からない

tensorflowのビギナーです
適当な面白そうな奴を見つけて走らせればどうか？とおもったので
https://github.com/vanhuyz/CycleGAN-TensorFlow
を走らせてみました。
これトレーニングのソースを見ると

python
1import tensorflow as tf
2from model import CycleGAN
3from reader import Reader
4from datetime import datetime
5import os
6import logging
7from utils import ImagePool
8
9FLAGS = tf.flags.FLAGS
10
11tf.flags.DEFINE_integer('batch_size', 1, 'batch size, default: 1')
12tf.flags.DEFINE_integer('image_size', 256, 'image size, default: 256')
13tf.flags.DEFINE_bool('use_lsgan', True,
14                     'use lsgan (mean squared error) or cross entropy loss, default: True')
15tf.flags.DEFINE_string('norm', 'instance',
16                       '[instance, batch] use instance norm or batch norm, default: instance')
17tf.flags.DEFINE_integer('lambda1', 10,
18                        'weight for forward cycle loss (X->Y->X), default: 10')
19tf.flags.DEFINE_integer('lambda2', 10,
20                        'weight for backward cycle loss (Y->X->Y), default: 10')
21tf.flags.DEFINE_float('learning_rate', 2e-4,
22                      'initial learning rate for Adam, default: 0.0002')
23tf.flags.DEFINE_float('beta1', 0.5,
24                      'momentum term of Adam, default: 0.5')
25tf.flags.DEFINE_float('pool_size', 50,
26                      'size of image buffer that stores previously generated images, default: 50')
27tf.flags.DEFINE_integer('ngf', 64,
28                        'number of gen filters in first conv layer, default: 64')
29
30tf.flags.DEFINE_string('X', 'data/tfrecords/apple.tfrecords',
31                       'X tfrecords file for training, default: data/tfrecords/apple.tfrecords')
32tf.flags.DEFINE_string('Y', 'data/tfrecords/orange.tfrecords',
33                       'Y tfrecords file for training, default: data/tfrecords/orange.tfrecords')
34tf.flags.DEFINE_string('load_model', None,
35                        'folder of saved model that you wish to continue training (e.g. 20170602-1936), default: None')
36
37
38def train():
39  if FLAGS.load_model is not None:
40    checkpoints_dir = "checkpoints/" + FLAGS.load_model.lstrip("checkpoints/")
41  else:
42    current_time = datetime.now().strftime("%Y%m%d-%H%M")
43    checkpoints_dir = "checkpoints/{}".format(current_time)
44    try:
45      os.makedirs(checkpoints_dir)
46    except os.error:
47      pass
48
49  graph = tf.Graph()
50  with graph.as_default():
51    cycle_gan = CycleGAN(
52        X_train_file=FLAGS.X,
53        Y_train_file=FLAGS.Y,
54        batch_size=FLAGS.batch_size,
55        image_size=FLAGS.image_size,
56        use_lsgan=FLAGS.use_lsgan,
57        norm=FLAGS.norm,
58        lambda1=FLAGS.lambda1,
59        lambda2=FLAGS.lambda2,
60        learning_rate=FLAGS.learning_rate,
61        beta1=FLAGS.beta1,
62        ngf=FLAGS.ngf
63    )
64    G_loss, D_Y_loss, F_loss, D_X_loss, fake_y, fake_x = cycle_gan.model()
65    optimizers = cycle_gan.optimize(G_loss, D_Y_loss, F_loss, D_X_loss)
66
67    summary_op = tf.summary.merge_all()
68    train_writer = tf.summary.FileWriter(checkpoints_dir, graph)
69    saver = tf.train.Saver()
70
71  with tf.Session(graph=graph) as sess:
72    if FLAGS.load_model is not None:
73      checkpoint = tf.train.get_checkpoint_state(checkpoints_dir)
74      meta_graph_path = checkpoint.model_checkpoint_path + ".meta"
75      restore = tf.train.import_meta_graph(meta_graph_path)
76      restore.restore(sess, tf.train.latest_checkpoint(checkpoints_dir))
77      step = int(meta_graph_path.split("-")[2].split(".")[0])
78    else:
79      sess.run(tf.global_variables_initializer())
80      step = 0
81
82    coord = tf.train.Coordinator()
83    threads = tf.train.start_queue_runners(sess=sess, coord=coord)
84
85    try:
86      fake_Y_pool = ImagePool(FLAGS.pool_size)
87      fake_X_pool = ImagePool(FLAGS.pool_size)
88
89      while not coord.should_stop():
90        # get previously generated images
91        fake_y_val, fake_x_val = sess.run([fake_y, fake_x])
92
93        # train
94        _, G_loss_val, D_Y_loss_val, F_loss_val, D_X_loss_val, summary = (
95              sess.run(
96                  [optimizers, G_loss, D_Y_loss, F_loss, D_X_loss, summary_op],
97                  feed_dict={cycle_gan.fake_y: fake_Y_pool.query(fake_y_val),
98                             cycle_gan.fake_x: fake_X_pool.query(fake_x_val)}
99              )
100        )
101
102        train_writer.add_summary(summary, step)
103        train_writer.flush()
104
105        if step % 100 == 0:
106          logging.info('-----------Step %d:-------------' % step)
107          logging.info('  G_loss   : {}'.format(G_loss_val))
108          logging.info('  D_Y_loss : {}'.format(D_Y_loss_val))
109          logging.info('  F_loss   : {}'.format(F_loss_val))
110          logging.info('  D_X_loss : {}'.format(D_X_loss_val))
111
112        if step % 10000 == 0:
113          save_path = saver.save(sess, checkpoints_dir + "/model.ckpt", global_step=step)
114          logging.info("Model saved in file: %s" % save_path)
115
116        step += 1
117
118    except KeyboardInterrupt:
119      logging.info('Interrupted')
120      coord.request_stop()
121    except Exception as e:
122      coord.request_stop(e)
123    finally:
124      save_path = saver.save(sess, checkpoints_dir + "/model.ckpt", global_step=step)
125      logging.info("Model saved in file: %s" % save_path)
126      # When done, ask the threads to stop.
127      coord.request_stop()
128      coord.join(threads)
129
130def main(unused_argv):
131  train()
132
133if __name__ == '__main__':
134  logging.basicConfig(level=logging.INFO)
135  tf.app.run()

こうなっています。
これをみるとＫＥＹの割り込み（ＣＴＬ＋Ｃ）か処理が例外のときに終了するらしいのはわかったのですがそれ以外にも停止することがあるのでしょうか？ってのがまず一個目の分からない分からないことです。
（このソースで割り込み以外で処理を終了させることができるのだろうか？ってことです。それともＴＦ独特の終わらせ方があるのか？）
次に
学習中に
logging.info('-----------Step %d:-------------' % step)
logging.info(' G_loss : {}'.format(G_loss_val))
logging.info(' D_Y_loss : {}'.format(D_Y_loss_val))
logging.info(' F_loss : {}'.format(F_loss_val))
logging.info(' D_X_loss : {}'.format(D_X_loss_val))
の値を出力しますが、
現在
INFO:root:-----------Step 22700:-------------
INFO:root: G_loss : 2.449359893798828
INFO:root: D_Y_loss : 0.11760234832763672
INFO:root: F_loss : 2.8169190883636475
INFO:root: D_X_loss : 0.05803598836064339
みたいな値ですがこれをどう評価すべきか分かりません。