使用tf.slim.evaluation.evaluate_once时的NotFoundError

Question

当我想使用slim.evaluation.evaluate_once()函数评估我的模型时，遇到了NotFoundError。它告诉我找不到模型的键或值。如下所示：

Running evaluation Loop...
INFO:tensorflow:Starting evaluation at 2017-08-25-11:40:57
INFO:tensorflow:Starting evaluation at 2017-08-25-11:40:57
INFO:tensorflow:Restoring parameters from tmp/flowers/finetune_log/model.ckpt-5000
INFO:tensorflow:Restoring parameters from tmp/flowers/finetune_log/model.ckpt-5000

NotFoundError                             Traceback (most recent call last)
/home/wangx/Dev_env/.tensorflow/lib/python3.5/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
   1326     try:
-> 1327       return fn(*args)
   1328     except errors.OpError as e:

/home/wangx/Dev_env/.tensorflow/lib/python3.5/site-packages/tensorflow/python/client/session.py in _run_fn(session, feed_dict, fetch_list, target_list, options, run_metadata)
   1305                                    feed_dict, fetch_list, target_list,
-> 1306                                    status, run_metadata)
   1307 

/usr/lib/python3.5/contextlib.py in __exit__(self, type, value, traceback)
     65             try:
---> 66                 next(self.gen)
     67             except StopIteration:

...
NotFoundError (see above for traceback): Key InceptionV1/Mixed_4c/Branch_0/Conv2d_0a_1x1/biases not found in checkpoint
     [[Node: save/RestoreV2_44 = RestoreV2[dtypes=[DT_FLOAT], _device="/job:localhost/replica:0/task:0/cpu:0"](_arg_save/Const_0_0, save/RestoreV2_44/tensor_names, save/RestoreV2_44/shape_and_slices)]]
     [[Node: save/RestoreV2_6/_1 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/gpu:0", send_device="/job:localhost/replica:0/task:0/cpu:0", send_device_incarnation=1, tensor_name="edge_238_save/RestoreV2_6", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/gpu:0"]()]]

我将我的检查点保存在./tmp/ flowers /finetune_log，然后在本教程之后下载flowers照片。我从培训中得到的检查点文件有问题吗？或者我在做评估的时候漏掉了什么？下面是我的评估代码：

from datasets import flowers
from nets import inception

with tf.Graph().as_default():
    tf.logging.set_verbosity(tf.logging.INFO)
    tf_global_step = slim.get_or_create_global_step()
    dataset = flowers.get_split('validation', 'tmp/flowers')
    images, labels = load_batch(dataset)
    logits, endpoints = inception.inception_v1(images, num_classes=dataset.num_classes, is_training=False)
    predictions =tf.argmax(logits, 1)

    # Define the metrics:
    names_to_values, names_to_updates = slim.metrics.aggregate_metric_map({
    'eval/Accuracy': slim.metrics.streaming_accuracy(predictions, labels),
    'eval/Recall': slim.metrics.streaming_recall(predictions, labels)})

    print('Running evaluation Loop...')
    checkpoint_path = tf.train.latest_checkpoint('tmp/flowers/finetune_log')

    metric_values = slim.evaluation.evaluate_once(
    num_evals=20,
    master='',
    checkpoint_path=checkpoint_path,
    logdir='tmp/flowers/eval_finetune_log',
    eval_op=names_to_updates.values(),
    final_op=names_to_values.values())

以防万一，下面是我的训练代码：

def get_init_fn():
    """Returns a function run by the chief worker to warm-start the training."""
    checkpoint_exclude_scopes=["InceptionV1/Logits", "InceptionV1/AuxLogits"]

    exclusions = [scope.strip() for scope in checkpoint_exclude_scopes]

    variables_to_restore = []
    for var in slim.get_model_variables():
        excluded = False
        for exclusion in exclusions:
            if var.op.name.startswith(exclusion):
                excluded = True
                break
        if not excluded:
            variables_to_restore.append(var)

    return slim.assign_from_checkpoint_fn(
      os.path.join('tmp/checkpoints', 'inception_v1.ckpt'),
      variables_to_restore)

train_dir = 'tmp/flowers/finetune_log'

with tf.Graph().as_default():
    dataset = flowers.get_split('train', 'tmp/flowers')
    images, labels = load_batch(dataset)

    with slim.arg_scope(inception.inception_v1_arg_scope()):
        logits, _ = inception.inception_v1(images, num_classes=dataset.num_classes, is_training=True)

    one_hot_labels = slim.one_hot_encoding(labels, 5)
    slim.losses.softmax_cross_entropy(logits, one_hot_labels)
    total_loss = slim.losses.get_total_loss()

    tf.summary.scalar('losses/Total Loss', total_loss)

    optimizer = tf.train.AdamOptimizer(learning_rate=0.01)
    train_op = slim.learning.create_train_op(total_loss, optimizer)

    final_loss = slim.learning.train(
        train_op,
        logdir=train_dir,
        init_fn=get_init_fn(),
        number_of_steps=5000,
        save_summaries_secs=1)

print('done.')

非常感谢。它阻碍了我很长时间。

Answer 1

我发现在评估片段中，如果我执行以下两个更改，程序就可以运行：

1. 为模型定义了slim.arg_scope()，我认为这就是NotFoundError信标程序不知道模型的参数的原因，所以代码应该像这样改变：

images, labels = load_batch(dataset)
with slim.arg_scope(inception.inception_v1_arg_scope()):
    logits, _ = inception.inception_v1(images, num_classes=dataset.num_classes, is_training=True)
predictions =tf.argmax(logits, 1)

2.我删除了slim.metrics.aggregate_metric_map()，并使用了一个简单的度量：

accuracy, accuracy_updates = slim.metrics.streaming_accuracy(predictions, labels)

现在它可以运行了。