Upgrade tensorflow to 2 (#73)

Author: xuhdev

Dockerfile

@@ -14,7 +14,7 @@
 # limitations under the License.
 #
 
-FROM quay.io/codait/max-base:v1.4.0
+FROM quay.io/codait/max-base:v1.5.0
 
 ARG model_bucket=https://max-cdn.cdn.appdomain.cloud/max-image-caption-generator/1.0.0
 ARG model_file=assets.tar.gz

api/predict.py

@@ -18,7 +18,7 @@
 from core.model import ModelWrapper
 
 from flask import abort
-from flask_restplus import fields
+from flask_restx import fields
 from werkzeug.datastructures import FileStorage
 
 

core/inference_utils/inference_wrapper_base.py

@@ -68,7 +68,7 @@ def build_model(self, model_config):
         Returns:
           model: The model object.
         """
-        tf.logging.fatal("Please implement build_model in subclass")
+        tf.compat.v1.logging.fatal("Please implement build_model in subclass")
 
     def _create_restore_fn(self, checkpoint_path, saver):
         """Creates a function that restores a model from checkpoint.
@@ -86,16 +86,16 @@ def _create_restore_fn(self, checkpoint_path, saver):
           ValueError: If checkpoint_path does not refer to a checkpoint file or a
             directory containing a checkpoint file.
         """
-        if tf.gfile.IsDirectory(checkpoint_path):
+        if tf.compat.v1.gfile.IsDirectory(checkpoint_path):
             checkpoint_path = tf.train.latest_checkpoint(checkpoint_path)
             if not checkpoint_path:
                 raise ValueError("No checkpoint file found in: %s" % checkpoint_path)
 
         def _restore_fn(sess):
-            tf.logging.info("Loading model from checkpoint: %s", checkpoint_path)
+            tf.compat.v1.logging.info("Loading model from checkpoint: %s", checkpoint_path)
             saver.restore(sess, checkpoint_path)
-            tf.logging.info("Successfully loaded checkpoint: %s",
-                            os.path.basename(checkpoint_path))
+            tf.compat.v1.logging.info("Successfully loaded checkpoint: %s",
+                                      os.path.basename(checkpoint_path))
 
         return _restore_fn
 
@@ -111,9 +111,9 @@ def build_graph_from_config(self, model_config, checkpoint_path):
           restore_fn: A function such that restore_fn(sess) loads model variables
             from the checkpoint file.
         """
-        tf.logging.info("Building model.")
+        tf.compat.v1.logging.info("Building model.")
         self.build_model(model_config)
-        saver = tf.train.Saver()
+        saver = tf.compat.v1.train.Saver()
 
         return self._create_restore_fn(checkpoint_path, saver)
 
@@ -132,18 +132,18 @@ def build_graph_from_proto(self, graph_def_file, saver_def_file,
             from the checkpoint file.
         """
         # Load the Graph.
-        tf.logging.info("Loading GraphDef from file: %s", graph_def_file)
+        tf.compat.v1.logging.info("Loading GraphDef from file: %s", graph_def_file)
         graph_def = tf.GraphDef()
-        with tf.gfile.FastGFile(graph_def_file, "rb") as f:
+        with tf.compat.v1.gfile.FastGFile(graph_def_file, "rb") as f:
             graph_def.ParseFromString(f.read())
         tf.import_graph_def(graph_def, name="")
 
         # Load the Saver.
-        tf.logging.info("Loading SaverDef from file: %s", saver_def_file)
-        saver_def = tf.train.SaverDef()
-        with tf.gfile.FastGFile(saver_def_file, "rb") as f:
+        tf.compat.v1.logging.info("Loading SaverDef from file: %s", saver_def_file)
+        saver_def = tf.compat.v1.train.SaverDef()
+        with tf.compat.v1.gfile.FastGFile(saver_def_file, "rb") as f:
             saver_def.ParseFromString(f.read())
-        saver = tf.train.Saver(saver_def=saver_def)
+        saver = tf.compat.v1.train.Saver(saver_def=saver_def)
 
         return self._create_restore_fn(checkpoint_path, saver)
 
@@ -159,7 +159,7 @@ def feed_image(self, sess, encoded_image):
         Returns:
           state: A numpy array of shape [1, state_size].
         """
-        tf.logging.fatal("Please implement feed_image in subclass")
+        tf.compat.v1.logging.fatal("Please implement feed_image in subclass")
 
     def inference_step(self, sess, input_feed, state_feed):
         """Runs one step of inference.
@@ -176,6 +176,6 @@ def inference_step(self, sess, input_feed, state_feed):
             current inference step (e.g. serialized numpy array containing
             activations from a particular model layer.).
         """
-        tf.logging.fatal("Please implement inference_step in subclass")
+        tf.compat.v1.logging.fatal("Please implement inference_step in subclass")
 
 # pylint: enable=unused-argument

core/inference_utils/vocabulary.py

@@ -39,11 +39,11 @@ def __init__(self,
           end_word: Special word denoting sentence end.
           unk_word: Special word denoting unknown words.
         """
-        if not tf.gfile.Exists(vocab_file):
-            tf.logging.fatal("Vocab file %s not found.", vocab_file)
-        tf.logging.info("Initializing vocabulary from file: %s", vocab_file)
+        if not tf.compat.v1.gfile.Exists(vocab_file):
+            tf.compat.v1.logging.fatal("Vocab file %s not found.", vocab_file)
+        tf.compat.v1.logging.info("Initializing vocabulary from file: %s", vocab_file)
 
-        with tf.gfile.GFile(vocab_file, mode="r") as f:
+        with tf.compat.v1.gfile.GFile(vocab_file, mode="r") as f:
             reverse_vocab = list(f.readlines())
         reverse_vocab = [line.split()[0] for line in reverse_vocab]
         if start_word not in reverse_vocab:
@@ -54,7 +54,7 @@ def __init__(self,
             reverse_vocab.append(unk_word)
         vocab = dict([(x, y) for (y, x) in enumerate(reverse_vocab)])
 
-        tf.logging.info("Created vocabulary with %d words" % len(vocab))
+        tf.compat.v1.logging.info("Created vocabulary with %d words" % len(vocab))
 
         self.vocab = vocab  # vocab[word] = id
         self.reverse_vocab = reverse_vocab  # reverse_vocab[id] = word

core/model.py

@@ -30,6 +30,8 @@
 
 logger = logging.getLogger()
 
+tf.compat.v1.disable_eager_execution()
+
 
 class ModelWrapper(MAXModelWrapper):
 
@@ -42,7 +44,7 @@ def __init__(self, path=DEFAULT_MODEL_PATH):
                                                        path)
         g.finalize()
         self.model = model
-        sess = tf.Session(graph=g)
+        sess = tf.compat.v1.Session(graph=g)
         # Load the model from checkpoint.
         restore_fn(sess)
         self.sess = sess

core/ops/image_embedding.py

@@ -21,9 +21,8 @@
 
 import tensorflow as tf
 
-from tensorflow.contrib.slim.python.slim.nets.inception_v3 import inception_v3_base
-
-slim = tf.contrib.slim
+import tf_slim as slim
+from tf_slim.nets.inception_v3 import inception_v3_base
 
 
 def inception_v3(images,
@@ -83,19 +82,19 @@ def inception_v3(images,
     else:
         weights_regularizer = None
 
-    with tf.variable_scope(scope, "InceptionV3", [images]) as scope:
+    with tf.compat.v1.variable_scope(scope, "InceptionV3", [images]) as scope:
         with slim.arg_scope(
                 [slim.conv2d, slim.fully_connected],
                 weights_regularizer=weights_regularizer,
                 trainable=trainable):
             with slim.arg_scope(
                     [slim.conv2d],
-                    weights_initializer=tf.truncated_normal_initializer(stddev=stddev),
+                    weights_initializer=tf.compat.v1.truncated_normal_initializer(stddev=stddev),
                     activation_fn=tf.nn.relu,
                     normalizer_fn=slim.batch_norm,
                     normalizer_params=batch_norm_params):
                 net, end_points = inception_v3_base(images, scope=scope)
-                with tf.variable_scope("logits"):
+                with tf.compat.v1.variable_scope("logits"):
                     shape = net.get_shape()
                     net = slim.avg_pool2d(net, shape[1:3], padding="VALID", scope="pool")
                     net = slim.dropout(
@@ -108,6 +107,6 @@ def inception_v3(images,
     # Add summaries.
     if add_summaries:
         for v in end_points.values():
-            tf.contrib.layers.summaries.summarize_activation(v)
+            slim.summarize_activation(v)
 
     return net

core/ops/image_processing.py

@@ -35,12 +35,12 @@ def distort_image(image, thread_id):
         [0, 1].
     """
     # Randomly flip horizontally.
-    with tf.name_scope("flip_horizontal", values=[image]):
+    with tf.name_scope("flip_horizontal"):
         image = tf.image.random_flip_left_right(image)
 
     # Randomly distort the colors based on thread id.
     color_ordering = thread_id % 2
-    with tf.name_scope("distort_color", values=[image]):
+    with tf.name_scope("distort_color"):
         if color_ordering == 0:
             image = tf.image.random_brightness(image, max_delta=32. / 255.)
             image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
@@ -95,7 +95,7 @@ def image_summary(name, image):
             tf.summary.image(name, tf.expand_dims(image, 0))
 
     # Decode image into a float32 Tensor of shape [?, ?, 3] with values in [0, 1).
-    with tf.name_scope("decode", values=[encoded_image]):
+    with tf.name_scope("decode"):
         if image_format == "jpeg":
             image = tf.image.decode_jpeg(encoded_image, channels=3)
         elif image_format == "png":
@@ -110,16 +110,16 @@ def image_summary(name, image):
         raise ValueError("Invalid resize parameters height: '{0}' width: '{1}'".format(resize_height, resize_width))
 
     if resize_height:
-        image = tf.image.resize_images(image,
-                                       size=[resize_height, resize_width],
-                                       method=tf.image.ResizeMethod.BILINEAR)
+        image = tf.image.resize(image,
+                                size=[resize_height, resize_width],
+                                method=tf.image.ResizeMethod.BILINEAR)
 
     # Crop to final dimensions.
     if is_training:
         image = tf.random_crop(image, [height, width, 3])
     else:
         # Central crop, assuming resize_height > height, resize_width > width.
-        image = tf.image.resize_image_with_crop_or_pad(image, height, width)
+        image = tf.image.resize_with_crop_or_pad(image, height, width)
 
     image_summary("resized_image", image)
 

core/ops/inputs.py

@@ -85,7 +85,7 @@ def prefetch_input_data(reader,
     """
     data_files = []
     for pattern in file_pattern.split(","):
-        data_files.extend(tf.gfile.Glob(pattern))
+        data_files.extend(tf.compat.v1.gfile.Glob(pattern))
     if not data_files:
         tf.logging.fatal("Found no input files matching %s", file_pattern)
     else:

core/show_and_tell_model.py

@@ -24,6 +24,7 @@
 from __future__ import print_function
 
 import tensorflow as tf
+import tf_slim as slim
 
 from core.ops import image_embedding
 from core.ops import image_processing
@@ -52,7 +53,7 @@ def __init__(self, config, mode, train_inception=False):
         self.train_inception = train_inception
 
         # Reader for the input data.
-        self.reader = tf.TFRecordReader()
+        self.reader = tf.compat.v1.TFRecordReader()
 
         # To match the "Show and Tell" paper we initialize all variables with a
         # random uniform initializer.
@@ -129,10 +130,10 @@ def build_inputs(self):
         """
         if self.mode == "inference":
             # In inference mode, images and inputs are fed via placeholders.
-            image_feed = tf.placeholder(dtype=tf.string, shape=[], name="image_feed")
-            input_feed = tf.placeholder(dtype=tf.int64,
-                                        shape=[None],  # batch_size
-                                        name="input_feed")
+            image_feed = tf.compat.v1.placeholder(dtype=tf.string, shape=[], name="image_feed")
+            input_feed = tf.compat.v1.placeholder(dtype=tf.int64,
+                                                  shape=[None],  # batch_size
+                                                  name="input_feed")
 
             # Process image and insert batch dimensions.
             images = tf.expand_dims(self.process_image(image_feed), 0)
@@ -192,12 +193,12 @@ def build_image_embeddings(self):
             self.images,
             trainable=self.train_inception,
             is_training=self.is_training())
-        self.inception_variables = tf.get_collection(
-            tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
+        self.inception_variables = tf.compat.v1.get_collection(
+            tf.compat.v1.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
 
         # Map inception output into embedding space.
-        with tf.variable_scope("image_embedding") as scope:
-            image_embeddings = tf.contrib.layers.fully_connected(
+        with tf.compat.v1.variable_scope("image_embedding") as scope:
+            image_embeddings = slim.layers.fully_connected(
                 inputs=inception_output,
                 num_outputs=self.config.embedding_size,
                 activation_fn=None,
@@ -219,8 +220,8 @@ def build_seq_embeddings(self):
         Outputs:
           self.seq_embeddings
         """
-        with tf.variable_scope("seq_embedding"), tf.device("/cpu:0"):
-            embedding_map = tf.get_variable(
+        with tf.compat.v1.variable_scope("seq_embedding"), tf.device("/cpu:0"):
+            embedding_map = tf.compat.v1.get_variable(
                 name="map",
                 shape=[self.config.vocab_size, self.config.embedding_size],
                 initializer=self.initializer)
@@ -245,15 +246,15 @@ def build_model(self):
         # This LSTM cell has biases and outputs tanh(new_c) * sigmoid(o), but the
         # modified LSTM in the "Show and Tell" paper has no biases and outputs
         # new_c * sigmoid(o).
-        lstm_cell = tf.contrib.rnn.BasicLSTMCell(
+        lstm_cell = tf.compat.v1.nn.rnn_cell.BasicLSTMCell(
             num_units=self.config.num_lstm_units, state_is_tuple=True)
         if self.mode == "train":
             lstm_cell = tf.contrib.rnn.DropoutWrapper(
                 lstm_cell,
                 input_keep_prob=self.config.lstm_dropout_keep_prob,
                 output_keep_prob=self.config.lstm_dropout_keep_prob)
 
-        with tf.variable_scope("lstm", initializer=self.initializer) as lstm_scope:
+        with tf.compat.v1.variable_scope("lstm", initializer=self.initializer) as lstm_scope:
             # Feed the image embeddings to set the initial LSTM state.
             zero_state = lstm_cell.zero_state(
                 batch_size=self.image_embeddings.get_shape()[0], dtype=tf.float32)
@@ -268,9 +269,9 @@ def build_model(self):
                 tf.concat(axis=1, values=initial_state, name="initial_state")
 
                 # Placeholder for feeding a batch of concatenated states.
-                state_feed = tf.placeholder(dtype=tf.float32,
-                                            shape=[None, sum(lstm_cell.state_size)],
-                                            name="state_feed")
+                state_feed = tf.compat.v1.placeholder(dtype=tf.float32,
+                                                      shape=[None, sum(lstm_cell.state_size)],
+                                                      name="state_feed")
                 state_tuple = tf.split(value=state_feed, num_or_size_splits=2, axis=1)
 
                 # Run a single LSTM step.
@@ -293,8 +294,8 @@ def build_model(self):
         # Stack batches vertically.
         lstm_outputs = tf.reshape(lstm_outputs, [-1, lstm_cell.output_size])
 
-        with tf.variable_scope("logits") as logits_scope:
-            logits = tf.contrib.layers.fully_connected(
+        with tf.compat.v1.variable_scope("logits") as logits_scope:
+            logits = slim.layers.fully_connected(
                 inputs=lstm_outputs,
                 num_outputs=self.config.vocab_size,
                 activation_fn=None,
@@ -341,11 +342,11 @@ def restore_fn(sess):
 
     def setup_global_step(self):
         """Sets up the global step Tensor."""
-        global_step = tf.Variable(
+        global_step = tf.compat.v1.Variable(
             initial_value=0,
             name="global_step",
             trainable=False,
-            collections=[tf.GraphKeys.GLOBAL_STEP, tf.GraphKeys.GLOBAL_VARIABLES])
+            collections=[tf.compat.v1.GraphKeys.GLOBAL_STEP, tf.compat.v1.GraphKeys.GLOBAL_VARIABLES])
 
         self.global_step = global_step
 

requirements.txt

@@ -1,3 +1,4 @@
-tensorflow==1.15.4
-Pillow==8.2.0
-numpy==1.17.4
+tensorflow==2.6.0
+tf_slim==1.1.0
+Pillow==8.3.1
+# NumPy version is dictated by tensorflow