Add spatial attention.

The spatial attention described in the paper was ported from the internal
implementation and is disabled by default.

attention_ocr/README.md

@@ -142,6 +142,9 @@ python train.py --dataset_name=newtextdataset
 
 Please note that eval.py will also require the same flag.
 
+To learn how to store a data in the FSNS
+ format please refer to the https://stackoverflow.com/a/44461910/743658.
+
 2. Define a new dataset format. The model needs the following data to train:
 
 - images: input images,  shape [batch_size x H x W x 3];
@@ -176,4 +179,4 @@ The main difference between this version and the version used in the paper - for
 the paper we used a distributed training with 50 GPU (K80) workers (asynchronous
 updates), the provided checkpoint was created using this code after ~6 days of
 training on a single GPU (Titan X) (it reached 81% after 24 hours of training),
-the coordinate encoding is missing TODO(alexgorban@).
+the coordinate encoding is disabled by default.

attention_ocr/python/model.py

@@ -55,6 +55,10 @@ SequenceLossParams = collections.namedtuple('SequenceLossParams', [
     'label_smoothing', 'ignore_nulls', 'average_across_timesteps'
 ])
 
+EncodeCoordinatesParams = collections.namedtuple('EncodeCoordinatesParams', [
+  'enabled'
+])
+
 
 def _dict_to_array(id_to_char, default_character):
   num_char_classes = max(id_to_char.keys()) + 1
@@ -162,7 +166,8 @@ class Model(object):
         SequenceLossParams(
             label_smoothing=0.1,
             ignore_nulls=True,
-            average_across_timesteps=False)
+            average_across_timesteps=False),
+        'encode_coordinates_fn': EncodeCoordinatesParams(enabled=False)
     }
 
   def set_mparam(self, function, **kwargs):
@@ -293,6 +298,30 @@ class Model(object):
     scores = tf.reshape(selected_scores, shape=(-1, self._params.seq_length))
     return ids, log_prob, scores
 
+  def encode_coordinates_fn(self, net):
+    """Adds one-hot encoding of coordinates to different views in the networks.
+
+    For each "pixel" of a feature map it adds a onehot encoded x and y
+    coordinates.
+
+    Args:
+      net: a tensor of shape=[batch_size, height, width, num_features]
+
+    Returns:
+      a tensor with the same height and width, but altered feature_size.
+    """
+    mparams = self._mparams['encode_coordinates_fn']
+    if mparams.enabled:
+      batch_size, h, w, _ = net.shape.as_list()
+      x, y = tf.meshgrid(tf.range(w), tf.range(h))
+      w_loc = slim.one_hot_encoding(x, num_classes=w)
+      h_loc = slim.one_hot_encoding(y, num_classes=h)
+      loc = tf.concat([h_loc, w_loc], 2)
+      loc = tf.tile(tf.expand_dims(loc, 0), [batch_size, 1, 1, 1])
+      return tf.concat([net, loc], 3)
+    else:
+      return net
+
   def create_base(self,
                   images,
                   labels_one_hot,
@@ -324,6 +353,9 @@ class Model(object):
       ]
       logging.debug('Conv tower: %s', nets[0])
 
+      nets = [self.encode_coordinates_fn(net) for net in nets]
+      logging.debug('Conv tower w/ encoded coordinates: %s', nets[0])
+
       net = self.pool_views_fn(nets)
       logging.debug('Pooled views: %s', net)
 

attention_ocr/python/model_test.py

@@ -62,8 +62,9 @@ class ModelTest(tf.test.TestCase):
         self.rng.randint(low=0, high=255,
                          size=self.images_shape).astype('float32'),
         name='input_node')
-    self.fake_conv_tower_np = tf.constant(
-        self.rng.randn(*self.conv_tower_shape).astype('float32'))
+    self.fake_conv_tower_np = self.rng.randn(
+        *self.conv_tower_shape).astype('float32')
+    self.fake_conv_tower = tf.constant(self.fake_conv_tower_np)
     self.fake_logits = tf.constant(
         self.rng.randn(*self.chars_logit_shape).astype('float32'))
     self.fake_labels = tf.constant(
@@ -162,6 +163,87 @@ class ModelTest(tf.test.TestCase):
     # This test checks that the loss function is 'runnable'.
     self.assertEqual(loss_np.shape, tuple())
 
+  def encode_coordinates_alt(self, net):
+    """An alternative implemenation for the encoding coordinates.
+
+    Args:
+      net: a tensor of shape=[batch_size, height, width, num_features]
+
+    Returns:
+      a list of tensors with encoded image coordinates in them.
+    """
+    batch_size, h, w, _ = net.shape.as_list()
+    h_loc = [
+      tf.tile(
+          tf.reshape(
+              tf.contrib.layers.one_hot_encoding(
+                  tf.constant([i]), num_classes=h), [h, 1]), [1, w])
+      for i in xrange(h)
+    ]
+    h_loc = tf.concat([tf.expand_dims(t, 2) for t in h_loc], 2)
+    w_loc = [
+      tf.tile(
+          tf.contrib.layers.one_hot_encoding(tf.constant([i]), num_classes=w),
+          [h, 1]) for i in xrange(w)
+    ]
+    w_loc = tf.concat([tf.expand_dims(t, 2) for t in w_loc], 2)
+    loc = tf.concat([h_loc, w_loc], 2)
+    loc = tf.tile(tf.expand_dims(loc, 0), [batch_size, 1, 1, 1])
+    return tf.concat([net, loc], 3)
+
+  def test_encoded_coordinates_have_correct_shape(self):
+    model = self.create_model()
+    model.set_mparam('encode_coordinates_fn', enabled=True)
+    conv_w_coords_tf = model.encode_coordinates_fn(self.fake_conv_tower)
+
+    with self.test_session() as sess:
+      conv_w_coords = sess.run(conv_w_coords_tf)
+
+    batch_size, height, width, feature_size = self.conv_tower_shape
+    self.assertEqual(conv_w_coords.shape, (batch_size, height, width,
+                                           feature_size + height + width))
+
+  def test_disabled_coordinate_encoding_returns_features_unchanged(self):
+    model = self.create_model()
+    model.set_mparam('encode_coordinates_fn', enabled=False)
+    conv_w_coords_tf = model.encode_coordinates_fn(self.fake_conv_tower)
+
+    with self.test_session() as sess:
+      conv_w_coords = sess.run(conv_w_coords_tf)
+
+    self.assertAllEqual(conv_w_coords, self.fake_conv_tower_np)
+
+  def test_coordinate_encoding_is_correct_for_simple_example(self):
+    shape = (1, 2, 3, 4)  # batch_size, height, width, feature_size
+    fake_conv_tower = tf.constant(2 * np.ones(shape), dtype=tf.float32)
+    model = self.create_model()
+    model.set_mparam('encode_coordinates_fn', enabled=True)
+    conv_w_coords_tf = model.encode_coordinates_fn(fake_conv_tower)
+
+    with self.test_session() as sess:
+      conv_w_coords = sess.run(conv_w_coords_tf)
+
+    # Original features
+    self.assertAllEqual(conv_w_coords[0, :, :, :4],
+                        [[[2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2]],
+                         [[2, 2, 2, 2], [2, 2, 2, 2], [2, 2, 2, 2]]])
+    # Encoded coordinates
+    self.assertAllEqual(conv_w_coords[0, :, :, 4:],
+                        [[[1, 0, 1, 0, 0], [1, 0, 0, 1, 0], [1, 0, 0, 0, 1]],
+                         [[0, 1, 1, 0, 0], [0, 1, 0, 1, 0], [0, 1, 0, 0, 1]]])
+
+  def test_alt_implementation_of_coordinate_encoding_returns_same_values(self):
+    model = self.create_model()
+    model.set_mparam('encode_coordinates_fn', enabled=True)
+    conv_w_coords_tf = model.encode_coordinates_fn(self.fake_conv_tower)
+    conv_w_coords_alt_tf = self.encode_coordinates_alt(self.fake_conv_tower)
+
+    with self.test_session() as sess:
+      conv_w_coords_tf, conv_w_coords_alt_tf = sess.run(
+          [conv_w_coords_tf, conv_w_coords_alt_tf])
+
+    self.assertAllEqual(conv_w_coords_tf, conv_w_coords_alt_tf)
+
 
 class CharsetMapperTest(tf.test.TestCase):
   def test_text_corresponds_to_ids(self):