TensorFlow cifar10_eval.py のオリジナル画像で試す。

結論を先に言ってしまえば、評価したい画像を　tf.read_file（）で読み込んで、
tf.image.decode_jpeg　等で加工するまでは良いのですが、

cifar10.inference()　で渡す際には、128個のバッチ形式 にしないとエラーになるようです。

おんちゃんが CIFAR-10 のサンプルをダウンロードしたのが、2017/12/29 だったのですが、
どうやら、参考にしたサイトの CIFAR-10版 とおんちゃんがダウンロードした版とで処理内容が
変わっているようです。

それに対応したコードは、こちらです。
#------------------- # 自前の画像の取り込み #------------------ f = tf.read_file('./20170430200754.jpg') my_image = tf.image.decode_jpeg(f, channels=3) my_image = tf.image.resize_images(my_image, (24, 24)) # 取り込んだ後の加工は、 cifar10_inputs.py の read_cifar10(), # _generate_image_and_label_batch() に習って加工します。 my_image = tf.reshape(my_image,[24,24,3]) # Subtract off the mean and divide by the variance of the pixels. my_image_float = tf.image.per_image_standardization(my_image) # Set the shapes of tensors. my_image_float.set_shape([24, 24, 3]) print("--- my_image_float ----") print(my_image_float) #-------------------------- # batch 形式に変換しないといけないみたい。(それも 128 サイズにするみたい) # sample from cifar10_input._generate_image_and_label_batch() batch_size=128 num_preprocess_threads = 1 # Ensure that the random shuffling has good mixing properties. min_fraction_of_examples_in_queue = 0.4 NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000 min_queue_examples = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN * min_fraction_of_examples_in_queue) num_threads=1 min_after_dequese = 1 #---------------- # バッチに変換 #---------------- print("evaluate_x(): #2.1") #print ('Filling queue with %d CIFAR images before starting to train. ' # 'This will take a few minutes.' % min_queue_examples) #images = tf.train.batch( # [my_image_float], # batch_size=batch_size, # num_threads=num_preprocess_threads, # capacity=min_queue_examples + 3 * batch_size) my_image_bat = tf.train.batch( [my_image_float], batch_size=batch_size, num_threads=num_threads, capacity=min_after_dequese + 3 * batch_size) print("evaluate_x(): #2.2") print('--- my_image_bat ---') print(my_image_bat) #tf.summary.image('images', [image]) # Build a Graph that computes the logits predictions from the # inference model. # ここで、グラフ(推論のモデル) を得ている by nishi logits = cifar10.inference(my_image_bat)

序に、具体的関数は、こちらです。
#################################### # eval_once_x() # 自前の画像の評価用で、最初の 1 件の評価結果を表示します。 ################################### def eval_once_x(saver, summary_writer, logits, summary_op): classes = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'] """Run Eval once. Args: saver: Saver. summary_writer: Summary writer. top_k_op: Top K op. summary_op: Summary op. """ with tf.Session() as sess: #print('\neval_once_x(): #1') ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir) if ckpt and ckpt.model_checkpoint_path: # Restores from checkpoint # ここでエラーみたい #print('\neval_once_x(): #2') #print(ckpt.model_checkpoint_path) saver.restore(sess, ckpt.model_checkpoint_path) # Assuming model_checkpoint_path looks something like: # /my-favorite-path/cifar10_train/model.ckpt-0, # extract global_step from it. global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1] else: print('No checkpoint file found') return # Start the queue runners. coord = tf.train.Coordinator() try: threads = [] for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS): threads.extend(qr.create_threads(sess, coord=coord, daemon=True, start=True)) #num_iter = int(math.ceil(FLAGS.num_examples / FLAGS.batch_size)) #true_count = 0 # Counts the number of correct predictions. #total_sample_count = num_iter * FLAGS.batch_size #step = 0 #while step < num_iter and not coord.should_stop(): # predictions = sess.run([top_k_op]) # true_count += np.sum(predictions) # step += 1 _summary_op, prediction = sess.run([summary_op, logits]) i = 0 num=1 for i in range(num): value = sess.run(tf.argmax(prediction[i], 0)) print("image: {0}/{1}".format((i+1), num)) #print('Correct: ', classes[correct[i]]) print('Prediction:', classes[value]) print(prediction[i], "\n") # Compute precision @ 1. #precision = true_count / total_sample_count #print('%s: precision @ 1 = %.3f' % (datetime.now(), precision)) summary = tf.Summary() #summary.ParseFromString(sess.run(summary_op)) summary.ParseFromString(_summary_op) #summary.value.add(tag='Precision @ 1', simple_value=precision) summary_writer.add_summary(summary, global_step) except Exception as e: # pylint: disable=broad-except coord.request_stop(e) coord.request_stop() coord.join(threads, stop_grace_period_secs=10) #################################### # evaluate_x(): # 自前の画像での評価 ##################################### def evaluate_x(): with tf.Graph().as_default() as g: # Get images and labels for CIFAR-10. eval_data = FLAGS.eval_data == 'test' #images0, labels0 = cifar10.inputs(eval_data=eval_data) print("evaluate_x(): #1") #print(images0) #------------------- # 自前の画像の取り込み #------------------ f = tf.read_file('./20170430200754.jpg') my_image = tf.image.decode_jpeg(f, channels=3) my_image = tf.image.resize_images(my_image, (24, 24)) # 取り込んだ後の加工は、 cifar10_inputs.py の read_cifar10(), # _generate_image_and_label_batch() に習って加工します。 my_image = tf.reshape(my_image,[24,24,3]) # Subtract off the mean and divide by the variance of the pixels. my_image_float = tf.image.per_image_standardization(my_image) # Set the shapes of tensors. my_image_float.set_shape([24, 24, 3]) print("--- my_image_float ----") print(my_image_float) #-------------------------- # batch 形式に変換しないといけないみたい。(それも 128 サイズにするみたい) # sample from cifar10_input._generate_image_and_label_batch() batch_size=128 num_preprocess_threads = 1 # Ensure that the random shuffling has good mixing properties. min_fraction_of_examples_in_queue = 0.4 NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000 min_queue_examples = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN * min_fraction_of_examples_in_queue) num_threads=1 min_after_dequese = 1 #---------------- # バッチに変換 #---------------- print("evaluate_x(): #2.1") #print ('Filling queue with %d CIFAR images before starting to train. ' # 'This will take a few minutes.' % min_queue_examples) #images = tf.train.batch( # [my_image_float], # batch_size=batch_size, # num_threads=num_preprocess_threads, # capacity=min_queue_examples + 3 * batch_size) my_image_bat = tf.train.batch( [my_image_float], batch_size=batch_size, num_threads=num_threads, capacity=min_after_dequese + 3 * batch_size) print("evaluate_x(): #2.2") print('--- my_image_bat ---') print(my_image_bat) #tf.summary.image('images', [image]) # Build a Graph that computes the logits predictions from the # inference model. # ここで、グラフ(推論のモデル) を得ている by nishi logits = cifar10.inference(my_image_bat) # Calculate predictions. #top_k_op = tf.nn.in_top_k(logits, labels, 1) # Restore the moving average version of the learned variables for eval. variable_averages = tf.train.ExponentialMovingAverage( cifar10.MOVING_AVERAGE_DECAY) variables_to_restore = variable_averages.variables_to_restore() #print('evaluate_x: #10') #print(variables_to_restore) # ここでエラー? saver = tf.train.Saver(variables_to_restore) summary_op = tf.summary.merge_all() summary_writer = tf.summary.FileWriter(FLAGS.eval_dir, g) eval_once_x(saver, summary_writer, logits, summary_op) #eval_once(saver, summary_writer, logits, summary_op, labels) def main(argv=None): # pylint: disable=unused-argument cifar10.maybe_download_and_extract() if tf.gfile.Exists(FLAGS.eval_dir): tf.gfile.DeleteRecursively(FLAGS.eval_dir) tf.gfile.MakeDirs(FLAGS.eval_dir) #evaluate() evaluate_x()

その後、cifar10.py で指定しているのが判かりました。
FLAGS.batch_size=128
cifar10_train.py を実行した時に、指定した値に合わせるみたいです。

# cifar10_train.py # Basic model parameters. #parser.add_argument('--batch_size', type=int, default=128, # help='Number of images to process in a batch.') parser.add_argument('--batch_size', type=int, default=1, help='Number of images to process in a batch.') parser.add_argument('--data_dir', type=str, default='/tmp/cifar10_data', help='Path to the CIFAR-10 data directory.') parser.add_argument('--use_fp16', type=bool, default=False, help='Train the model using fp16.') FLAGS = parser.parse_args()

H29年の暮れから、H30年の正月３ヶ日は、これで潰れました。
やれやれです。おんちゃんのH30年の正月もやっと迎えれます。

by おんちゃん 2018.1.3 

後書き 1)
実際に試してみると、saver = tf.train.Saver(variables_to_restore) エラーになりました。
InvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [18,384] rhs shape= [2304,384] [[Node: save/Assign_5 = Assign[T=DT_FLOAT, _class=["loc:@local3/weights"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](local3/weights, save/RestoreV2_5)]]
なにか、Restore した形式と、合致していない云々...のエラーのような?

後でこのログをよくよく見てみると、
local3/weights
とあるので、これは、
cifar10.py の inference() の中の、local3 の定義だと何と無く、見て取れます。

cifar10.py line 247 # local3 with tf.variable_scope('local3') as scope: # Move everything into depth so we can perform a single matrix multiply. reshape = tf.reshape(pool2, [FLAGS.batch_size, -1]) dim = reshape.get_shape()[1].value weights = _variable_with_weight_decay('weights', shape=[dim, 384], stddev=0.04, wd=0.004) biases = _variable_on_cpu('biases', [384], tf.constant_initializer(0.1)) local3 = tf.nn.relu(tf.matmul(reshape, weights) + biases, name=scope.name) _activation_summary(local3)

結局、該当箇所を、下記の様に変更すれば、1個の画像でもOKとなりました。
cifar10.py line 247 # local3 with tf.variable_scope('local3') as scope: # Move everything into depth so we can perform a single matrix multiply. #reshape = tf.reshape(pool2, [FLAGS.batch_size, -1]) # changed by nishi reshape = tf.reshape(pool2, [-1, 2303]) #print('reshape = ',reshape) dim = reshape.get_shape()[1].value weights = _variable_with_weight_decay('weights', shape=[dim, 384], stddev=0.04, wd=0.004) biases = _variable_on_cpu('biases', [384], tf.constant_initializer(0.1)) local3 = tf.nn.relu(tf.matmul(reshape, weights) + biases, name=scope.name) _activation_summary(local3)

そうすると、結局、tf.expand_dims() で先頭に1次元追加の、擬似バッチで、OKになりした。
def evaluate_x(): . . . #------------------- # 自前の画像の取り込み #------------------ f = tf.read_file('./20170430200754.jpg') my_image = tf.image.decode_jpeg(f, channels=3) my_image = tf.image.resize_images(my_image, (24, 24)) # 取り込んだ後の加工は、 cifar10_inputs.py の read_cifar10(), # _generate_image_and_label_batch() に習って加工します。 my_image = tf.reshape(my_image,[24,24,3]) # Subtract off the mean and divide by the variance of the pixels. my_image_float = tf.image.per_image_standardization(my_image) # Set the shapes of tensors. my_image_float.set_shape([24, 24, 3]) print("--- my_image_float ----") print(my_image_float) # tf.expand_dims() で、先頭に1次元追加した、擬似バッチにします。 # create a fake batch of images (batch_size = 1) my_image_float_x = tf.expand_dims(my_image_float, 0) logits = cifar10.inference(my_image_float_x) . . . </p>