def _read_word_record(data_queue):
reader = tf.TFRecordReader() # Construct a general reader
key, example_serialized = reader.read(data_queue)
feature_map = {
'image/encoded': tf.FixedLenFeature( [], dtype=tf.string,
default_value='' ),
'image/labels': tf.VarLenFeature( dtype=tf.int64 ),
'image/width': tf.FixedLenFeature( [1], dtype=tf.int64,
default_value=1 ),
'image/filename': tf.FixedLenFeature([], dtype=tf.string,
default_value='' ),
'text/string': tf.FixedLenFeature([], dtype=tf.string,
default_value='' ),
'text/length': tf.FixedLenFeature( [1], dtype=tf.int64,
default_value=1 )
}
features = tf.parse_single_example( example_serialized, feature_map )
image = tf.image.decode_jpeg( features['image/encoded'], channels=1 ) #gray
width = tf.cast( features['image/width'], tf.int32) # for ctc_loss
label = tf.serialize_sparse( features['image/labels'] ) # for batching
length = features['text/length']
text = features['text/string']
filename = features['image/filename']
return image,width,label,length,text,filename
def testDeserializeFailsInconsistentRank(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
sp_input1 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
input1_val = self._SparseTensorValue_1x1x1()
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(
serialized_concat, dtype=tf.int32)
with self.assertRaisesOpError(
r"Inconsistent rank across SparseTensors: rank prior to "
r"SparseTensor\[1\] was: 3 but rank of SparseTensor\[1\] is: 4"):
sess.run(
sp_deserialized, {sp_input0: input0_val, sp_input1: input1_val})
def testDeserializeFailsWrongType(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
sp_input1 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
input1_val = self._SparseTensorValue_3x4(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(
serialized_concat, dtype=tf.int64)
with self.assertRaisesOpError(
r"Requested SparseTensor of type int64 but "
r"SparseTensor\[0\].values.dtype\(\) == int32"):
sess.run(
sp_deserialized, {sp_input0: input0_val, sp_input1: input1_val})
def testSerializeSparseTensor(self):
sp_input = tf.SparseTensor(
indices=tf.constant([[1]], dtype=tf.int64),
values=tf.constant([2], dtype=tf.int64),
dense_shape=[2])
with self.cached_session():
serialized_sp = tf.serialize_sparse(sp_input, 'serialize_name', tf.string)
self.assertEqual((3,), serialized_sp.shape)
self.assertTrue(serialized_sp[0].numpy()) # check non-empty
def testDeserializeFailsWrongType(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
sp_input1 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
input1_val = self._SparseTensorValue_3x4(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(serialized_concat,
dtype=tf.int64)
with self.assertRaisesOpError(
r"Requested SparseTensor of type int64 but "
r"SparseTensor\[0\].values.dtype\(\) == int32"):
sess.run(sp_deserialized, {
sp_input0: input0_val,
sp_input1: input1_val
})
def testSerializeDeserializeMany(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorValue_5x6(np.arange(6))
sp_input1 = self._SparseTensorValue_3x4(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(
serialized_concat, dtype=tf.int32)
combined_indices, combined_values, combined_shape = sess.run(
sp_deserialized)
self.assertAllEqual(combined_indices[:6, 0], [0] * 6) # minibatch 0
self.assertAllEqual(combined_indices[:6, 1:], sp_input0[0])
self.assertAllEqual(combined_indices[6:, 0], [1] * 6) # minibatch 1
self.assertAllEqual(combined_indices[6:, 1:], sp_input1[0])
self.assertAllEqual(combined_values[:6], sp_input0[1])
self.assertAllEqual(combined_values[6:], sp_input1[1])
self.assertAllEqual(combined_shape, [2, 5, 6])
def testDeserializeFailsInconsistentRank(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
sp_input1 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
input1_val = self._SparseTensorValue_1x1x1()
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(serialized_concat,
dtype=tf.int32)
with self.assertRaisesOpError(
r"Inconsistent rank across SparseTensors: rank prior to "
r"SparseTensor\[1\] was: 3 but rank of SparseTensor\[1\] is: 4"
):
sess.run(sp_deserialized, {
sp_input0: input0_val,
sp_input1: input1_val
})
def testSerializeDeserializeMany(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorValue_5x6(np.arange(6))
sp_input1 = self._SparseTensorValue_3x4(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = tf.serialize_sparse(sp_input1)
serialized_concat = tf.stack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(
serialized_concat, dtype=tf.int32)
combined_indices, combined_values, combined_shape = sess.run(
sp_deserialized)
self.assertAllEqual(combined_indices[:6, 0], [0] * 6) # minibatch 0
self.assertAllEqual(combined_indices[:6, 1:], sp_input0[0])
self.assertAllEqual(combined_indices[6:, 0], [1] * 6) # minibatch 1
self.assertAllEqual(combined_indices[6:, 1:], sp_input1[0])
self.assertAllEqual(combined_values[:6], sp_input0[1])
self.assertAllEqual(combined_values[6:], sp_input1[1])
self.assertAllEqual(combined_shape, [2, 5, 6])
def _ReadExamples(filename_queue, shape, using_ctc, reader=None):
"""Builds network input tensor ops for TF Example.
Args:
filename_queue: Queue of filenames, from tf.train_bkp.string_input_producer
shape: ImageShape with the desired shape of the input.
using_ctc: Take the unpadded_class labels instead of padded.
reader: Function that returns an actual reader to read Examples from
input files. If None, uses tf.TFRecordReader().
Returns:
image: Float Tensor containing the input image scaled to [-1.28, 1.27].
height: Tensor int64 containing the height of the image.
width: Tensor int64 containing the width of the image.
labels: Serialized SparseTensor containing the int64 labels.
text: Tensor string of the utf8 truth text.
"""
if reader:
reader = reader()
else:
reader = tf.TFRecordReader()
_, example_serialized = reader.read(filename_queue)
example_serialized = tf.reshape(example_serialized, shape=[])
features = tf.parse_single_example(
example_serialized, {
'image/encoded':
parsing_ops.FixedLenFeature([1], dtype=tf.string,
default_value=''),
'image/text':
parsing_ops.FixedLenFeature([1], dtype=tf.string,
default_value=''),
'image/class':
parsing_ops.VarLenFeature(dtype=tf.int64),
'image/unpadded_class':
parsing_ops.VarLenFeature(dtype=tf.int64),
'image/height':
parsing_ops.FixedLenFeature([1], dtype=tf.int64, default_value=1),
'image/width':
parsing_ops.FixedLenFeature([1], dtype=tf.int64, default_value=1)
})
if using_ctc:
labels = features['image/unpadded_class']
else:
labels = features['image/class']
labels = tf.serialize_sparse(labels)
image = tf.reshape(features['image/encoded'], shape=[], name='encoded')
image = _ImageProcessing(image, shape)
height = tf.reshape(features['image/height'], [-1])
width = tf.reshape(features['image/width'], [-1])
text = tf.reshape(features['image/text'], shape=[])
return image, height, width, labels, text
def testDeserializeFailsInvalidProto(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = ["a", "b", "c"]
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(
serialized_concat, dtype=tf.int32)
with self.assertRaisesOpError(
r"Could not parse serialized_sparse\[1, 0\]"):
sess.run(sp_deserialized, {sp_input0: input0_val})
def testDeserializeFailsInvalidProto(self):
with self.test_session(use_gpu=False) as sess:
sp_input0 = self._SparseTensorPlaceholder()
input0_val = self._SparseTensorValue_5x6(np.arange(6))
serialized0 = tf.serialize_sparse(sp_input0)
serialized1 = ["a", "b", "c"]
serialized_concat = tf.pack([serialized0, serialized1])
sp_deserialized = tf.deserialize_many_sparse(serialized_concat,
dtype=tf.int32)
with self.assertRaisesOpError(
r"Could not parse serialized_sparse\[1, 0\]"):
sess.run(sp_deserialized, {sp_input0: input0_val})
def _ReadExamples(filename_queue, shape, using_ctc, reader=None):
"""Builds network input tensor ops for TF Example.
Args:
filename_queue: Queue of filenames, from tf.train.string_input_producer
shape: ImageShape with the desired shape of the input.
using_ctc: Take the unpadded_class labels instead of padded.
reader: Function that returns an actual reader to read Examples from
input files. If None, uses tf.TFRecordReader().
Returns:
image: Float Tensor containing the input image scaled to [-1.28, 1.27].
height: Tensor int64 containing the height of the image.
width: Tensor int64 containing the width of the image.
labels: Serialized SparseTensor containing the int64 labels.
text: Tensor string of the utf8 truth text.
"""
if reader:
reader = reader()
else:
reader = tf.TFRecordReader()
_, example_serialized = reader.read(filename_queue)
example_serialized = tf.reshape(example_serialized, shape=[])
features = tf.parse_single_example(
example_serialized,
{'image/encoded': parsing_ops.FixedLenFeature(
[1], dtype=tf.string, default_value=''),
'image/text': parsing_ops.FixedLenFeature(
[1], dtype=tf.string, default_value=''),
'image/class': parsing_ops.VarLenFeature(dtype=tf.int64),
'image/unpadded_class': parsing_ops.VarLenFeature(dtype=tf.int64),
'image/height': parsing_ops.FixedLenFeature(
[1], dtype=tf.int64, default_value=1),
'image/width': parsing_ops.FixedLenFeature(
[1], dtype=tf.int64, default_value=1)})
if using_ctc:
labels = features['image/unpadded_class']
else:
labels = features['image/class']
labels = tf.serialize_sparse(labels)
image = tf.reshape(features['image/encoded'], shape=[], name='encoded')
image = _ImageProcessing(image, shape)
height = tf.reshape(features['image/height'], [-1])
width = tf.reshape(features['image/width'], [-1])
text = tf.reshape(features['image/text'], shape=[])
return image, height, width, labels, text
def batchReadTf(tfpath):
with tf.Session() as sess:
# Create a list of filenames and pass it to a queue
filename_queue = tf.train.string_input_producer([tfpath], num_epochs=1)
# Define a reader and read the next record
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
# Decode the record read by the reader
features = tf.parse_single_example(serialized_example, features=feature_map)
image = tf.image.decode_jpeg(features['image/encoded'], channels=1) # gray
width = tf.cast(features['image/width'], tf.int32) # for ctc_loss
label = tf.serialize_sparse(features['image/labels']) # for batching
length = features['text/length']
text = features['text/string']
filename = features['image/filename']
# print text
init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
sess.run(init_op)
# Create a coordinator and run all QueueRunner objects
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
for batch_index in range(20):
img, lbl = sess.run([image, text])
print lbl,img.shape
img = img.reshape((img.shape[0:2]))
pimg = Image.fromarray(img, "L")
# pimg.save("../img/%d-%s.jpg" % (batch_index,lbl))
# img = img.astype(np.uint8)
# for j in range(6):
# plt.subplot(2, 3, j + 1)
# plt.imshow(img[j, ...]) example.features.feature['image/encoded']
# plt.title('cat' if lbl[j] == 0 else 'dog')
# plt.show()
# Stop the threads
coord.request_stop()
# Wait for threads to stop
coord.join(threads)
sess.close()
def _parser(self, serialized, distort=False):
tensor_dict = ops.parse_sequence_example(serialized)
tensor_dict['image/data'] = self._preproc(tensor_dict['image/data'])
(tensor_dict['image/data'],
tensor_dict['image/seq/char/bbox'],
tensor_dict['image/seq/line/bbox']) = preproc.pad_borders_or_shrink(
tensor_dict['image/data'], tensor_dict['image/seq/char/bbox'],
tensor_dict['image/seq/line/bbox'], self.pad_shape)
if self.sparse_labels:
tensor_dict['image/seq/char/id_sparse'] = tf.serialize_sparse(
ops.sparsify_label(tensor_dict['image/seq/char/id'],
tensor_dict['image/char/count'])
)
# if distort: image = distort_image(image)
self.feat_keys, features = zip(*tensor_dict.items())
return features
def txt_parse_fn(self, line, dataset_name):
"""
"""
line = tf.strings.split([line], sep='\t', maxsplit=1)
line = tf.sparse.to_dense(line, default_value='')
filename = tf.reshape(line[0][0], [])
text = tf.reshape(line[0][1], [])
length = tf.reshape(tf.py_func(text_length, [text], tf.int64), [-1])
text = tf.py_func(self.join_with_delimiter, [text], tf.string)
text = tf.strings.join([text, EOS_TOKEN], separator='')
image_content = tf.read_file(filename)
image = tf.image.decode_image(image_content,
channels=self.img_channels)
image.set_shape([None, None, self.img_channels])
label = self.indexing_fn(text, length, DELIMITER)
label = tf.serialize_sparse(label)
return image, label, length, text, filename, dataset_name
def serialize_sparse_tensor(self, sparse_tensor):
return tf.serialize_tensor(tf.serialize_sparse(sparse_tensor)).numpy()
