def main(_):
# if len(sys.argv) < 2 or sys.argv[-1].startswith('-'):
# print('Usage: mnist_export.py [--training_iteration=x] '
# '[--model_version=y] export_dir')
# sys.exit(-1)
if FLAGS.training_iteration <= 0:
print('Please specify a positive value for training iteration.')
sys.exit(-1)
if FLAGS.model_version <= 0:
print('Please specify a positive value for version number.')
sys.exit(-1)
print('Training...')
mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)
sess = tf.InteractiveSession()
serialized_tf_example = tf.placeholder(tf.string, name='tf_example')
feature_configs = {'x': tf.FixedLenFeature(shape=[784], dtype=tf.float32),} # 28*28 =784
tf_example = tf.parse_example(serialized_tf_example, feature_configs)
x = tf.identity(tf_example['x'], name='x')
y_ = tf.placeholder('float', shape=[None, 10])
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])
x_image = tf.reshape(x, [-1, 28, 28, 1])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])
h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
y = tf.nn.softmax(y_conv, name='y')
cross_entropy = -tf.reduce_sum(y_ * tf.log(y_conv))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
values, indices = tf.nn.top_k(y_conv, 10) # 从小到大排名每个分数以及相应的index
prediction_classes = tf.contrib.lookup.index_to_string(
tf.to_int64(indices),
mapping=tf.constant([str(i) for i in range(10)]))
sess.run(tf.global_variables_initializer())
for _ in range(FLAGS.training_iteration):
batch = mnist.train.next_batch(50)
train_step.run(feed_dict={x: batch[0],
y_: batch[1], keep_prob: 0.5})
correct_prediction = tf.equal(tf.argmax(y_conv, 1),
tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print('training accuracy %g' % accuracy.eval(feed_dict={
x: mnist.test.images,
y_: mnist.test.labels, keep_prob: 1.0}))
print('training is finished!')
export_path_base = FLAGS.export_path #"export_base" #sys.argv[-1]
export_path = os.path.join(compat.as_bytes(export_path_base),compat.as_bytes(str(FLAGS.model_version)))
print('Exporting trained model to', export_path)
if os.path.exists(export_path_base):
shutil.rmtree(export_path_base,ignore_errors=True)
builder = saved_model_builder.SavedModelBuilder(export_path)
# 分类的签名
classification_inputs = utils.build_tensor_info(serialized_tf_example)
classification_outputs_classes = utils.build_tensor_info(prediction_classes) # 预测的top 10的类别
classification_outputs_scores = utils.build_tensor_info(values) # 预测的top 10的分数
classification_signature = signature_def_utils.build_signature_def(
inputs={signature_constants.CLASSIFY_INPUTS:classification_inputs},
# 注意：此处的输出有两个值
outputs={
signature_constants.CLASSIFY_OUTPUT_CLASSES:
classification_outputs_classes,
signature_constants.CLASSIFY_OUTPUT_SCORES:
classification_outputs_scores
},
method_name=signature_constants.CLASSIFY_METHOD_NAME
)
# 预测的签名
tensor_info_x = utils.build_tensor_info(x)
tensor_info_y = utils.build_tensor_info(y_conv) #预测时不需要再计算softmax，直接比较就行
prediction_signature = signature_def_utils.build_signature_def(
inputs={'images': tensor_info_x},
outputs={'scores': tensor_info_y}, # 预测成为不同label的分数
method_name=signature_constants.PREDICT_METHOD_NAME)
legacy_init_op = tf.group(tf.initialize_all_tables(), name='legacy_init_op')
builder.add_meta_graph_and_variables(
sess, [tag_constants.SERVING],
signature_def_map={
'predict_images': prediction_signature,
"class_signature": classification_signature,
#signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: classification_signature, # serving_default
},
legacy_init_op=legacy_init_op)
builder.save()
print('new model exported!')
def parse_example_proto(example_serialized):
"""Parses an Example proto containing a training example of an image.
The output of the build_image_data.py image preprocessing script is a dataset
containing serialized Example protocol buffers. Each Example proto contains
the following fields:
image/height: 462
image/width: 581
image/colorspace: 'RGB'
image/channels: 3
image/class/label: 615
image/class/synset: 'n03623198'
image/class/text: 'knee pad'
image/object/bbox/xmin: 0.1
image/object/bbox/xmax: 0.9
image/object/bbox/ymin: 0.2
image/object/bbox/ymax: 0.6
image/object/bbox/label: 615
image/format: 'JPEG'
image/filename: 'ILSVRC2012_val_00041207.JPEG'
image/encoded: <JPEG encoded string>
Args:
example_serialized: scalar Tensor tf.string containing a serialized
Example protocol buffer.
Returns:
image_buffer: Tensor tf.string containing the contents of a JPEG file.
label: Tensor tf.int32 containing the label.
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged as
[ymin, xmin, ymax, xmax].
text: Tensor tf.string containing the human-readable label.
"""
# Dense features in Example proto.
feature_map = {
'image/encoded': tf.FixedLenFeature([], dtype=tf.string,
default_value=''),
'image/class/label': tf.FixedLenFeature([1], dtype=tf.int64,
default_value=-1),
'image/class/text': tf.FixedLenFeature([], dtype=tf.string,
default_value=''),
}
sparse_float32 = tf.VarLenFeature(dtype=tf.float32)
# Sparse features in Example proto.
feature_map.update(
{k: sparse_float32 for k in ['image/object/bbox/xmin',
'image/object/bbox/ymin',
'image/object/bbox/xmax',
'image/object/bbox/ymax']})
features = tf.parse_single_example(example_serialized, feature_map)
label = tf.cast(features['image/class/label'], dtype=tf.int32)
xmin = tf.expand_dims(features['image/object/bbox/xmin'].values, 0)
ymin = tf.expand_dims(features['image/object/bbox/ymin'].values, 0)
xmax = tf.expand_dims(features['image/object/bbox/xmax'].values, 0)
ymax = tf.expand_dims(features['image/object/bbox/ymax'].values, 0)
# Note that we impose an ordering of (y, x) just to make life difficult.
bbox = tf.concat(axis=0, values=[ymin, xmin, ymax, xmax])
# Force the variable number of bounding boxes into the shape
# [1, num_boxes, coords].
bbox = tf.expand_dims(bbox, 0)
bbox = tf.transpose(bbox, [0, 2, 1])
return features['image/encoded'], label, bbox, features['image/class/text']
# -*- coding: utf-8 -*-
# @Author : JaeZheng
# @Time : 2019/9/4 15:09
# @File : FileQueue_read.py
import tensorflow as tf
# tf.train.match_filenames_once获取文件列表
files = tf.train.match_filenames_once("./data/tfrecords-*")
# tf.train.string_input_producer创建输入文件的列表，shuffle设置是否打乱
file_queue = tf.train.string_input_producer(files, shuffle=False)
reader = tf.TFRecordReader()
_, serialized_example = reader.read(file_queue)
features = tf.parse_single_example(serialized_example,
features={
'i': tf.FixedLenFeature([], tf.int64),
'j': tf.FixedLenFeature([], tf.int64),
})
with tf.Session() as sess:
tf.local_variables_initializer().run()
print(sess.run(files))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for i in range(6):
print(sess.run([features['i'], features['j']]))
coord.request_stop()
coord.join(threads)
def get_dataset(dataset_name, split_name, dataset_dir):
"""Gets an instance of slim Dataset.
Args:
dataset_name: Dataset name.
split_name: A train/val Split name.
dataset_dir: The directory of the dataset sources.
Returns:
An instance of slim Dataset.
Raises:
ValueError: if the dataset_name or split_name is not recognized.
"""
if dataset_name not in _DATASETS_INFORMATION:
raise ValueError('The specified dataset is not supported yet.')
splits_to_sizes = _DATASETS_INFORMATION[dataset_name].splits_to_sizes
if split_name not in splits_to_sizes:
raise ValueError('data split name %s not recognized' % split_name)
# Prepare the variables for different datasets.
num_classes = _DATASETS_INFORMATION[dataset_name].num_classes
ignore_label = _DATASETS_INFORMATION[dataset_name].ignore_label
file_pattern = _FILE_PATTERN
file_pattern = os.path.join(dataset_dir, file_pattern % split_name)
# Specify how the TF-Examples are decoded.
keys_to_features = {
'image/encoded': tf.FixedLenFeature(
(), tf.string, default_value=''),
'image/filename': tf.FixedLenFeature(
(), tf.string, default_value=''),
'image/format': tf.FixedLenFeature(
(), tf.string, default_value='jpeg'),
'image/height': tf.FixedLenFeature(
(), tf.int64, default_value=0),
'image/width': tf.FixedLenFeature(
(), tf.int64, default_value=0),
'image/segmentation/class/encoded': tf.FixedLenFeature(
(), tf.string, default_value=''),
'image/segmentation/class/format': tf.FixedLenFeature(
(), tf.string, default_value='png'),
}
items_to_handlers = {
'image': tfexample_decoder.Image(
image_key='image/encoded',
format_key='image/format',
channels=3),
'image_name': tfexample_decoder.Tensor('image/filename'),
'height': tfexample_decoder.Tensor('image/height'),
'width': tfexample_decoder.Tensor('image/width'),
'labels_class': tfexample_decoder.Image(
image_key='image/segmentation/class/encoded',
format_key='image/segmentation/class/format',
channels=1),
}
decoder = tfexample_decoder.TFExampleDecoder(
keys_to_features, items_to_handlers)
return dataset.Dataset(
data_sources=file_pattern,
reader=tf.TFRecordReader,
decoder=decoder,
num_samples=splits_to_sizes[split_name],
items_to_descriptions=_ITEMS_TO_DESCRIPTIONS,
ignore_label=ignore_label,
num_classes=num_classes,
name=dataset_name,
multi_label=True)
def arbitrary_style_image_inputs(style_dataset_file,
batch_size=None,
image_size=None,
center_crop=True,
shuffle=True,
augment_style_images=False,
random_style_image_size=False,
min_rand_image_size=128,
max_rand_image_size=300):
"""Loads a batch of random style image given the path of tfrecord dataset.
This method does not return pre-compute Gram matrices for the images like
style_image_inputs. But it can provide data augmentation. If
augment_style_images is equal to True, then style images will randomly
modified (eg. changes in brightness, hue or saturation) for data
augmentation. If random_style_image_size is set to True then all images
in one batch will be resized to a random size.
Args:
style_dataset_file: str, path to the tfrecord dataset of style files.
batch_size: int. If provided, batches style images. Defaults to None.
image_size: int. The images will be resized bilinearly so that the smallest
side has size image_size. Defaults to None.
center_crop: bool. If True, center-crops to [image_size, image_size].
Defaults to False.
shuffle: bool, whether to shuffle style files at random. Defaults to False.
augment_style_images: bool. Wheather to augment style images or not.
random_style_image_size: bool. If this value is True, then all the style
images in one batch will be resized to a random size between
min_rand_image_size and max_rand_image_size.
min_rand_image_size: int. If random_style_image_size is True, this value
specifies the minimum image size.
max_rand_image_size: int. If random_style_image_size is True, this value
specifies the maximum image size.
Returns:
4-D tensor of shape [1, ?, ?, 3] with values in [0, 1] for the style
image (with random changes for data augmentation if
augment_style_image_size is set to true), and 0-D tensor for the style
label, 4-D tensor of shape [1, ?, ?, 3] with values in [0, 1] for the style
image without random changes for data augmentation.
Raises:
ValueError: if center cropping is requested but no image size is provided,
or if batch size is specified but center-cropping or
augment-style-images is not requested,
or if both augment-style-images and center-cropping are requested.
"""
if center_crop and image_size is None:
raise ValueError('center-cropping requires specifying the image size.')
if center_crop and augment_style_images:
raise ValueError(
'When augment_style_images is true images will be randomly cropped.'
)
if batch_size is not None and not center_crop and not augment_style_images:
raise ValueError(
'batching requires same image sizes (Set center-cropping or '
'augment_style_images to true)')
with tf.name_scope('style_image_processing'):
# Force all input processing onto CPU in order to reserve the GPU for the
# forward inference and back-propagation.
with tf.device('/cpu:0'):
filename_queue = tf.train.string_input_producer(
[style_dataset_file],
shuffle=False,
capacity=1,
name='filename_queue')
if shuffle:
examples_queue = tf.RandomShuffleQueue(
capacity=64,
min_after_dequeue=32,
dtypes=[tf.string],
name='random_examples_queue')
else:
examples_queue = tf.FIFOQueue(capacity=64,
dtypes=[tf.string],
name='fifo_examples_queue')
reader = tf.TFRecordReader()
_, value = reader.read(filename_queue)
enqueue_ops = [examples_queue.enqueue([value])]
tf.train.queue_runner.add_queue_runner(
tf.train.queue_runner.QueueRunner(examples_queue, enqueue_ops))
example_serialized = examples_queue.dequeue()
features = tf.parse_single_example(
example_serialized,
features={
'label': tf.FixedLenFeature([], tf.int64),
'image_raw': tf.FixedLenFeature([], tf.string)
})
image = tf.image.decode_jpeg(features['image_raw'])
image.set_shape([None, None, 3])
label = features['label']
if image_size is not None:
image_channels = image.shape[2].value
if augment_style_images:
image_orig = image
image = tf.image.random_brightness(image, max_delta=0.8)
image = tf.image.random_saturation(image,
lower=0.5,
upper=1.5)
image = tf.image.random_hue(image, max_delta=0.2)
image = tf.image.random_flip_left_right(image)
image = tf.image.random_flip_up_down(image)
random_larger_image_size = random_ops.random_uniform(
[],
minval=image_size + 2,
maxval=image_size + 200,
dtype=dtypes.int32)
image = _aspect_preserving_resize(
image, random_larger_image_size)
image = tf.random_crop(
image, size=[image_size, image_size, image_channels])
image.set_shape([image_size, image_size, image_channels])
image_orig = _aspect_preserving_resize(
image_orig, image_size + 2)
image_orig = _central_crop([image_orig], image_size,
image_size)[0]
image_orig.set_shape([image_size, image_size, 3])
elif center_crop:
image = _aspect_preserving_resize(image, image_size + 2)
image = _central_crop([image], image_size, image_size)[0]
image.set_shape([image_size, image_size, image_channels])
image_orig = image
else:
image = _aspect_preserving_resize(image, image_size)
image_orig = image
image = tf.to_float(image) / 255.0
image_orig = tf.to_float(image_orig) / 255.0
if batch_size is None:
image = tf.expand_dims(image, 0)
else:
[image, image_orig,
label] = tf.train.batch([image, image_orig, label],
batch_size=batch_size)
if random_style_image_size:
# Selects a random size for the style images and resizes all the images
# in the batch to that size.
image = _aspect_preserving_resize(
image,
random_ops.random_uniform([],
minval=min_rand_image_size,
maxval=max_rand_image_size,
dtype=dtypes.int32))
return image, label, image_orig
print(tf.__version__)
print("Finish parameter setting")
filename = "/planes_scannet_val.tfrecords"
filename_queue = tf.train.string_input_producer(['/planes_scannet_val.tfrecords'], num_epochs=1)
reader = tf.TFRecordReader()
print(filename_queue)
_, serialized_example = reader.read(filename_queue)
print(serialized_example)
features = tf.parse_single_example(
serialized_example,
# Defaults are not specified since both keys are required.
features={
#'height': tf.FixedLenFeature([], tf.int64),
#'width': tf.FixedLenFeature([], tf.int64),
'image_raw': tf.FixedLenFeature([], tf.string),
'image_path': tf.FixedLenFeature([], tf.string),
'num_planes': tf.FixedLenFeature([], tf.int64),
'plane': tf.FixedLenFeature([NUM_PLANES * 3], tf.float32),
#'plane_relation': tf.FixedLenFeature([NUM_PLANES * NUM_PLANES], tf.float32),
'segmentation_raw': tf.FixedLenFeature([], tf.string),
'depth': tf.FixedLenFeature([HEIGHT * WIDTH], tf.float32),
'normal': tf.FixedLenFeature([HEIGHT * WIDTH * 3], tf.float32),
'semantics_raw': tf.FixedLenFeature([], tf.string),
'boundary_raw': tf.FixedLenFeature([], tf.string),
'info': tf.FixedLenFeature([4 * 4 + 4], tf.float32),
})
sess=tf.Session()
print("stuck after this:1")
sess.run(tf.global_variables_initializer())
def load_examples():
if a.mode == 'train':
filename_queue = tf.train.string_input_producer([a.train_tfrecord])
elif a.mode == 'test':
filename_queue = tf.train.string_input_producer([a.test_tfrecord])
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(serialized_example,
features={
'im_mul_raw': tf.FixedLenFeature([], tf.string),
'im_blur_raw': tf.FixedLenFeature([], tf.string),
'im_lr_raw': tf.FixedLenFeature([], tf.string),
'im_ndvi_raw': tf.FixedLenFeature([], tf.string),
'im_ndwi_raw': tf.FixedLenFeature([], tf.string),
'im_pan_raw': tf.FixedLenFeature([], tf.string),
})
im_mul_1_raw = tf.decode_raw(features['im_mul_raw'], tf.uint8)
im_mul_1_raw = tf.reshape(im_mul_1_raw, [128, 128, 4])
im_mul_1_raw = tf.cast(im_mul_1_raw, tf.float32)
im_lr_1_raw = tf.decode_raw(features['im_lr_raw'], tf.uint8)
im_lr_1_raw = tf.reshape(im_lr_1_raw, [32, 32, 4])
im_lr_1_raw = tf.cast(im_lr_1_raw, tf.float32)
im_ndvi_raw = tf.decode_raw(features['im_ndvi_raw'], tf.uint8)
im_ndvi_raw = tf.reshape(im_ndvi_raw, [128, 128, 1])
im_ndvi_raw = tf.cast(im_ndvi_raw, tf.float32)
im_ndwi_raw = tf.decode_raw(features['im_ndwi_raw'], tf.uint8)
im_ndwi_raw = tf.reshape(im_ndwi_raw, [128, 128, 1])
im_ndwi_raw = tf.cast(im_ndwi_raw, tf.float32)
im_pan_raw = tf.decode_raw(features['im_pan_raw'], tf.uint8)
im_pan_raw = tf.reshape(im_pan_raw, [128, 128, 1])
im_pan_raw = tf.cast(im_pan_raw, tf.float32)
im_blur_1_raw = tf.decode_raw(features['im_blur_raw'], tf.uint8)
im_blur_1_raw = tf.reshape(im_blur_1_raw, [128, 128, 4])
im_blur_1_raw = tf.cast(im_blur_1_raw, tf.float32)
if a.mode == 'train':
inputs1_batch, inputs2_batch, inputs3_batch, inputs4_batch, targets_batch, inputs5_batch = tf.train.shuffle_batch(
[im_blur_1_raw, im_ndvi_raw, im_ndwi_raw, im_pan_raw, im_mul_1_raw, im_lr_1_raw],
batch_size=a.batch_size, capacity=200,
min_after_dequeue=100)
steps_per_epoch = int(a.train_count / a.batch_size)
elif a.mode == 'test':
inputs1_batch, inputs2_batch, inputs3_batch, inputs4_batch, targets_batch, inputs5_batch = tf.train.batch(
[im_blur_1_raw, im_ndvi_raw, im_ndwi_raw, im_pan_raw, im_mul_1_raw, im_lr_1_raw],
batch_size=a.batch_size, capacity=200)
steps_per_epoch = int(a.test_count / a.batch_size)
return Examples(
inputs1=inputs1_batch,
inputs2=inputs2_batch,
inputs3=inputs3_batch,
inputs4=inputs4_batch,
inputs5=inputs5_batch,
targets=targets_batch,
steps_per_epoch=steps_per_epoch,
)
import tensorflow as tf
reader = tf.TFRecordReader()
filename_queue = tf.train.string_input_producer(["./records/output.tfrecords"])
# 从文件中读出一个样例
_, serialized_example = reader.read(filename_queue)
# 解析读入的一个样例
features = tf.parse_single_example(
serialized_example,
features={
'image_raw': tf.FixedLenFeature([], tf.string),
'pixels': tf.FixedLenFeature([], tf.int64),
'label': tf.FixedLenFeature([], tf.int64),
})
images = tf.decode_raw(features['image_raw'], tf.uint8)
labels = tf.cast(features['label'], tf.int32)
pixels = tf.cast(features['pixels'], tf.int32)
sess = tf.Session()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for i in range(10):
inage, label, pixel = sess.run([images, labels, pixels])
def parser(record):
# preprocess "inp_perm" and "tgt_perm"
def _process_perm_feature(example, prefix):
for b in range(len(bin_sizes)):
cnt = example.pop("{}_cnt_{}".format(prefix, b))[0]
tup = example.pop("{}_tup_{}".format(prefix, b))
tup = tf.reshape(tf.sparse_tensor_to_dense(tup),
shape=[cnt, 2])
# tf.float32
perm = tf.sparse_to_dense(
sparse_indices=tup,
output_shape=[tgt_len, bin_sizes[b]],
sparse_values=1.0,
default_value=0.0)
example["{}_perm_{}".format(prefix, b)] = perm
# whether allow the last batch with a potentially shorter length
if use_tpu:
record_spec = {
"inputs": tf.FixedLenFeature([tgt_len], tf.int64),
"labels": tf.FixedLenFeature([tgt_len], tf.int64),
}
else:
record_spec = {
"inputs": tf.VarLenFeature(tf.int64),
"labels": tf.VarLenFeature(tf.int64),
}
# permutation related features
if bin_sizes and use_tpu:
# tf.float32
record_spec["inp_mask"] = tf.FixedLenFeature([tgt_len],
tf.float32)
record_spec["tgt_mask"] = tf.FixedLenFeature([tgt_len],
tf.float32)
record_spec["head_labels"] = tf.FixedLenFeature([tgt_len],
tf.int64)
for b in range(len(bin_sizes)):
record_spec["inp_cnt_{}".format(b)] = tf.FixedLenFeature(
[1], tf.int64)
record_spec["inp_tup_{}".format(b)] = tf.VarLenFeature(
tf.int64)
record_spec["tgt_cnt_{}".format(b)] = tf.FixedLenFeature(
[1], tf.int64)
record_spec["tgt_tup_{}".format(b)] = tf.VarLenFeature(
tf.int64)
# retrieve serialized example
example = tf.parse_single_example(serialized=record,
features=record_spec)
# transform permutation tuples to permutation matrices
if bin_sizes and use_tpu:
_process_perm_feature(example, "inp")
_process_perm_feature(example, "tgt")
# cast int64 into int32
# cast sparse to dense
for key in list(example.keys()):
val = example[key]
if tf.keras.backend.is_sparse(val):
val = tf.sparse.to_dense(val)
if val.dtype == tf.int64:
val = tf.to_int32(val)
example[key] = val
if use_tpu:
return example
else:
return example["inputs"], example["labels"]
def _multi_read_and_decode(filename_queue):
"""Reads a multi record and converts it to a dictionary of tensors.
Args:
filename_queue: Tensor Queue, list of input files.
Returns:
Dictionary of tensors containing the information from a multi record.
Image height.
"""
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(
serialized_example,
# Defaults are not specified since both keys are required.
features={
'height': tf.FixedLenFeature([], tf.int64),
'width': tf.FixedLenFeature([], tf.int64),
'depth': tf.FixedLenFeature([], tf.int64),
'label_1': tf.FixedLenFeature([], tf.int64),
'label_2': tf.FixedLenFeature([], tf.int64),
'image_raw_1': tf.FixedLenFeature([], tf.string),
'image_raw_2': tf.FixedLenFeature([], tf.string),
'merged_raw': tf.FixedLenFeature([], tf.string),
})
# Decode 3 images
features['image_raw_1'] = tf.decode_raw(features['image_raw_1'], tf.uint8)
features['image_raw_2'] = tf.decode_raw(features['image_raw_2'], tf.uint8)
features['merged_raw'] = tf.decode_raw(features['merged_raw'], tf.uint8)
image_pixels = 36 * 36
features['image_raw_1'].set_shape([image_pixels])
features['image_raw_2'].set_shape([image_pixels])
features['merged_raw'].set_shape([image_pixels])
# Convert from [0, 255] -> [-0.5, 0.5] floats.
features['image_raw_1'] = tf.cast(features['image_raw_1'],
tf.float32) * (1. / 255)
features['image_raw_2'] = tf.cast(features['image_raw_2'],
tf.float32) * (1. / 255)
features['merged_raw'] = tf.cast(features['merged_raw'],
tf.float32) * (1. / 255)
# Convert label from a scalar uint8 tensor to an int32 scalar.
features['label_1'] = tf.cast(features['label_1'], tf.int32)
features['label_2'] = tf.cast(features['label_2'], tf.int32)
# Convert the dictionary to the format used in the code.
res_features = {}
res_features['height'] = features['height']
res_features['depth'] = features['depth']
res_features['images'] = features['merged_raw']
res_features['labels'] = tf.one_hot(features['label_1'], 10) + tf.one_hot(
features['label_2'], 10)
res_features['recons_label'] = features['label_1']
res_features['recons_image'] = features['image_raw_1']
res_features['spare_label'] = features['label_2']
res_features['spare_image'] = features['image_raw_2']
return res_features, 36
def _read_and_decode(filename_queue,
image_dim=28,
distort=False,
split='train',
evaluate=False):
"""Reads a single record and converts it to a tensor.
Args:
filename_queue: Tensor Queue, list of input files.
image_dim: Scalar, the height (and width) of the image in pixels.
distort: Boolean, whether to distort the input or not.
split: String, the split of the data (test or train) to read from.
evaluate: If set, avoid distorting the image.
Returns:
Dictionary of the (Image, label) and the image height.
"""
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(serialized_example,
features={
'image_raw':
tf.FixedLenFeature([], tf.string),
'label':
tf.FixedLenFeature([], tf.int64),
'height':
tf.FixedLenFeature([], tf.int64),
'width':
tf.FixedLenFeature([], tf.int64),
'depth':
tf.FixedLenFeature([], tf.int64)
})
# Convert from a scalar string tensor (whose single string has
# length image_pixel*image_pixel) to a uint8 tensor with shape
# [image_pixel, image_pixel, 1].
image = tf.decode_raw(features['image_raw'], tf.uint8)
image = tf.reshape(image, [image_dim, image_dim, 1])
image.set_shape([image_dim, image_dim, 1])
# Convert from [0, 255] -> [-0.5, 0.5] floats.
image = tf.cast(image, tf.float32) * (1. / 255)
if distort:
cropped_dim = image_dim - 4
if not evaluate:
image = tf.reshape(image, [image_dim, image_dim])
image = tf.random_crop(image, [cropped_dim, cropped_dim])
# 0.26179938779 is 15 degress in radians
image = tf.contrib.image.rotate(
image, random.uniform(-0.26179938779, 0.26179938779))
image = tf.reshape(image, [cropped_dim, cropped_dim, 1])
image.set_shape([cropped_dim, cropped_dim, 1])
else:
fraction = cropped_dim / image_dim
image = tf.image.central_crop(image, central_fraction=fraction)
image.set_shape([cropped_dim, cropped_dim, 1])
image_dim = cropped_dim
# Convert label from a scalar uint8 tensor to an int32 scalar.
label = tf.cast(features['label'], tf.int32)
features = {
'images': image,
'labels': tf.one_hot(label, 10),
'recons_image': image,
'recons_label': label,
}
return features, image_dim
import tensorflow as tf
import glob
data_path = '../datasets/coco/2017_training/tfrecords/' # address to save the hdf5 file
# data_path = '/data/cvg/lukas/datasets/coco/2017_test/version/v2/final/'
data_path = '/data/cvg/lukas/datasets/coco/2017_training/version/v6/tmp/'
data_path = '/data/cvg/lukas/datasets/coco/2017_training/version/v6/final/'
data_path = '/data/cvg/lukas/datasets/coco/2017_test/version/v3/final/'
data_path = '/home/lz01a008/git/yad2k/YAD2K/voc_conversion_scripts/VOCdevkit/tfrecords/train/'
name_image_feature = 'image/encoded'
name_image_feature = 'encoded'
with tf.Session() as sess:
feature = {name_image_feature: tf.FixedLenFeature([], tf.string)}
all_files = glob.glob(data_path + '*')
all_files = all_files if len(all_files) > 0 else [data_path]
print('counting in %s files...' % (len(all_files)))
# Create a list of filenames and pass it to a queue
filename_queue = tf.train.string_input_producer(all_files, 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)
#image = tf.decode_raw(features['image/encoded'], tf.uint8)
print(features)
image = features[name_image_feature]
def main(_):
assert FLAGS.tfrecord, '`tfrecord` is missing.'
filename = [FLAGS.tfrecord]
number = number_of_jpeg(filename)
filename_queue = tf.train.string_input_producer(filename) #Read tfrecord
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(serialized_example,
features={
'image/height':
tf.FixedLenFeature([], tf.int64),
'image/width':
tf.FixedLenFeature([], tf.int64),
'image/object/bbox/xmin':
tf.VarLenFeature(tf.float32),
'image/object/bbox/xmax':
tf.VarLenFeature(tf.float32),
'image/object/bbox/ymin':
tf.VarLenFeature(tf.float32),
'image/object/bbox/ymax':
tf.VarLenFeature(tf.float32),
'image/object/class/label':
tf.VarLenFeature(tf.int64),
'image/encoded':
tf.FixedLenFeature([], tf.string),
})
image = tf.image.decode_jpeg(features['image/encoded'])
#label = tf.cast(features['image/object/class/label'], tf.int32)
if not os.path.exists(FLAGS.output_filepath):
os.makedirs(FLAGS.output_filepath)
if FLAGS.pdf:
pp = PdfPages(FLAGS.output_filepath + '/out.pdf')
category_index = None
if FLAGS.pbtxt:
category_index = label_map_util.create_category_index_from_labelmap(
FLAGS.pbtxt)
print(category_index)
with tf.Session() as sess:
init_op = tf.initialize_all_variables()
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
t_h = tf.cast(features['image/height'], tf.int32)
t_w = tf.cast(features['image/width'], tf.int32)
#single_shape = tf.parallel_stack([1])
d_label = tf.sparse_tensor_to_dense(
features['image/object/class/label'], default_value=0)
d_xm = tf.sparse_tensor_to_dense(features['image/object/bbox/xmin'],
default_value=0)
d_xM = tf.sparse_tensor_to_dense(features['image/object/bbox/xmax'],
default_value=0)
d_ym = tf.sparse_tensor_to_dense(features['image/object/bbox/ymin'],
default_value=0)
d_yM = tf.sparse_tensor_to_dense(features['image/object/bbox/ymax'],
default_value=0)
#t_xm = tf.cast(features['image/object/bbox/xmin'], [tf.float32])
#t_xM = tf.cast(features['image/object/bbox/xmax'], [tf.float32])
#t_ym = tf.cast(features['image/object/bbox/ymin'], [tf.float32])
#t_yM = tf.cast(features['image/object/bbox/ymax'], [tf.float32])
t_l = tf.reshape(d_label, [-1])
t_xm = tf.reshape(d_xm, [-1])
t_xM = tf.reshape(d_xM, [-1])
t_ym = tf.reshape(d_ym, [-1])
t_yM = tf.reshape(d_yM, [-1])
boxid = 0
#axarr.autoscale(True)
for i in range(number):
#image = tf.reshape(image, [w, h, 3])
example, ls, w, h, xm, xM, ym, yM = sess.run(
[image, d_label, t_w, t_h, t_xm, t_xM, t_ym, t_yM])
img_data_jpg = tf.image.convert_image_dtype(example,
dtype=tf.uint8)
encode_image_jpg = tf.image.encode_jpeg(img_data_jpg)
folder = FLAGS.output_filepath
bx = []
bl = []
label = []
for idx, l in enumerate(ls):
#if l not in label:
#label.append(str(l))
bx.append([ym[idx], xm[idx], yM[idx], xM[idx]])
#bl.append([str(l)])
labelname = str(l)
if category_index:
labeldic = category_index[int(l)]
labelname = labeldic['name']
bl.append([labelname])
if labelname not in label:
label.append(labelname)
newname=folder+"/"+str(boxid)+'['+str(w)+'x'+str(h)+']'+'Label_'+labelname+'@[' \
+str(int(round(xm[idx]*w)))+','+str(int(round(ym[idx]*h)))+']to[' \
+str(int(round(xM[idx]*w)))+','+str(int(round(yM[idx]*h)))+'].jpg'
with tf.gfile.GFile(newname, 'wb') as f:
f.write(encode_image_jpg.eval())
boxid += 1
if FLAGS.pdf:
num = len(label)
with Image.open(newname) as im:
shape = len(np.array(im).shape)
#print(shape)
for idx, l in enumerate(bl):
lid = label.index(l[0])
if shape == 3:
gb = int(lid * 255.0 / num)
c = (255, 255 - gb, gb)
elif shape == 2:
c = int(255 - (lid * 128.0 / num))
#print(bx[idx])
#print(bl[idx])
#print(c)
vis_util.draw_bounding_boxes_on_image(
im,
np.array([bx[idx]]),
color=c,
display_str_list_list=[bl[idx]])
plt.imshow(im)
pp.savefig()
print(str(bl[idx]) + ":" + str(w) + "x" + str(h))
#input('press return to continue')
if FLAGS.pdf:
pp.close()
coord.request_stop()
coord.join(threads)
def parse_tfrecord_tf(self, record):
features = tf.parse_single_example(record, features={
'shape': tf.FixedLenFeature([3], tf.int64),
'data': tf.FixedLenFeature([], tf.string)})
data = tf.decode_raw(features['data'], tf.as_dtype(self.dtype))
return tf.reshape(data, features['shape'])
def main(_):
with tf.device(
tf.train.replica_device_setter(worker_device="/job:%s/task:%d" %
(FLAGS.job_name, FLAGS.task_index),
ps_device="/job:ps/cpu:0",
cluster=cluster)):
#
# 1. read training data
#
# image - 784 (=28 x 28) elements of grey-scaled integer value [0, 1]
# label - digit (0, 1, ..., 9)
train_queue = tf.train.string_input_producer(
[FLAGS.train_file],
num_epochs=200) # when all data is read, it raises OutOfRange
train_reader = tf.TFRecordReader()
_, train_serialized_exam = train_reader.read(train_queue)
train_exam = tf.parse_single_example(train_serialized_exam,
features={
'image_raw':
tf.FixedLenFeature([],
tf.string),
'label':
tf.FixedLenFeature([],
tf.int64)
})
train_image = tf.decode_raw(train_exam['image_raw'], tf.uint8)
train_image.set_shape([784])
train_image = tf.cast(train_image, tf.float32) * (1. / 255)
train_label = tf.cast(train_exam['label'], tf.int32)
train_batch_image, train_batch_label = tf.train.batch(
[train_image, train_label], batch_size=batch_size)
# # for debugging... (check input)
# with tf.Session() as sess:
# sess.run(tf.global_variables_initializer())
# tf.train.start_queue_runners(sess=sess)
# for i in range(2):
# debug_image, debug_label = sess.run([train_batch_image, train_batch_label])
# tf.summary.image('images', debug_image)
# print(debug_label)
#
# 2. define graph
#
# define input
plchd_image = tf.placeholder(
dtype=tf.float32, shape=(batch_size, 784)
) # here we use fixed dimension with batch_size. (Please use undefined dimension with None in production.)
plchd_label = tf.placeholder(
dtype=tf.int32, shape=(batch_size)
) # here we use fixed dimension with batch_size. (Please use undefined dimension with None in production.)
# define network and inference
# (simple 2 fully connected hidden layer : 784->128->64->10)
with tf.name_scope('hidden1'):
weights = tf.Variable(tf.truncated_normal([784, 128],
stddev=1.0 /
math.sqrt(float(784))),
name='weights')
biases = tf.Variable(tf.zeros([128]), name='biases')
hidden1 = tf.nn.relu(tf.matmul(plchd_image, weights) + biases)
with tf.name_scope('hidden2'):
weights = tf.Variable(tf.truncated_normal([128, 64],
stddev=1.0 /
math.sqrt(float(128))),
name='weights')
biases = tf.Variable(tf.zeros([64]), name='biases')
hidden2 = tf.nn.relu(tf.matmul(hidden1, weights) + biases)
with tf.name_scope('softmax_linear'):
weights = tf.Variable(tf.truncated_normal([64, 10],
stddev=1.0 /
math.sqrt(float(64))),
name='weights')
biases = tf.Variable(tf.zeros([10]), name='biases')
logits = tf.matmul(hidden2, weights) + biases
global_step = tf.train.get_or_create_global_step()
# define optimization (training)
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.07)
loss = tf.losses.sparse_softmax_cross_entropy(labels=plchd_label,
logits=logits)
train_op = optimizer.minimize(loss=loss, global_step=global_step)
#
# 3. run session
#
with tf.train.MonitoredTrainingSession(
master=server.target,
checkpoint_dir=FLAGS.out_dir,
is_chief=is_chief,
) as sess: # use tf.train.MonitoredTrainingSession for more advanced features ...
# train !!!
local_step_value = 0
array_image, array_label = sess.run(
[train_batch_image, train_batch_label])
while not sess.should_stop():
feed_dict = {
plchd_image: array_image,
plchd_label: array_label
}
_, global_step_value, loss_value, array_image, array_label = sess.run(
[
train_op, global_step, loss, train_batch_image,
train_batch_label
],
feed_dict=feed_dict)
local_step_value += 1
if local_step_value % 100 == 0:
print(
"Worker: Local Step %d, Global Step %d (Loss: %.2f)" %
(local_step_value, global_step_value, loss_value))
print('Training done !')
def slim_get_split(file_pattern='{}_????'):
# Features in Pascal VOC TFRecords.
keys_to_features = {
'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
'image/format': tf.FixedLenFeature((), tf.string,
default_value='jpeg'),
'image/image_id': tf.FixedLenFeature([1], tf.int64),
'image/height': tf.FixedLenFeature([1], tf.int64),
'image/width': tf.FixedLenFeature([1], tf.int64),
'image/channels': tf.FixedLenFeature([1], tf.int64),
'image/shape': tf.FixedLenFeature([3], tf.int64),
'image/object/bbox/xmin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/xmax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/label': tf.VarLenFeature(dtype=tf.int64),
}
items_to_handlers = {
'image':
slim.tfexample_decoder.Image('image/encoded', 'image/format'),
'image_id':
slim.tfexample_decoder.Tensor('image/image_id'),
'shape':
slim.tfexample_decoder.Tensor('image/shape'),
'object/bbox':
slim.tfexample_decoder.BoundingBox(['ymin', 'xmin', 'ymax', 'xmax'],
'image/object/bbox/'),
'object/label':
slim.tfexample_decoder.Tensor('image/object/bbox/label'),
}
decoder = slim.tfexample_decoder.TFExampleDecoder(keys_to_features,
items_to_handlers)
dataset = slim.dataset.Dataset(data_sources=file_pattern,
reader=tf.TFRecordReader,
decoder=decoder,
num_samples=100,
items_to_descriptions=None,
num_classes=81,
labels_to_names=None)
with tf.name_scope('dataset_data_provider'):
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=32,
common_queue_min=8,
shuffle=True,
num_epochs=1)
[org_image, image_id, shape, glabels_raw, gbboxes_raw] = provider.get(
['image', 'image_id', 'shape', 'object/label', 'object/bbox'])
image, glabels, gbboxes = ssd_preprocessing.preprocess_image(
org_image,
glabels_raw,
gbboxes_raw, [304, 304],
is_training=True,
data_format='channels_last',
output_rgb=False)
# image = tf.transpose(image, perm=(1, 2, 0))
save_image_op = tf.py_func(save_image_with_bbox, [
ssd_preprocessing.unwhiten_image(image),
tf.clip_by_value(glabels, 0, tf.int64.max),
tf.ones_like(glabels), gbboxes
],
tf.int64,
stateful=True)
# out_shape = [304, 304]
# image = tf.to_float(org_image)
# image = tf.image.resize_images(image, out_shape, method=tf.image.ResizeMethod.BILINEAR, align_corners=False)
# image.set_shape(out_shape + [3])
# save_image_op = tf.py_func(save_image_with_bbox, [image, glabels_raw, tf.ones_like(glabels_raw), gbboxes_raw], tf.int64, stateful=True)
"""
============before ssd_preprocessing===============:
image: Tensor("dataset_data_provider/case/cond/Merge:0", shape=(?, ?, 3), dtype=uint8)
glabels_raw: Tensor("dataset_data_provider/SparseToDense:0", shape=(?,), dtype=int64)
gbboxes_raw: Tensor("dataset_data_provider/transpose:0", shape=(?, 4), dtype=float32)
============after ssd_preprocessing================:
image: Tensor("ssd_preprocessing_train/merge_bgr:0", shape=(304, 304, 3), dtype=float32)
glabels_raw: Tensor("ssd_preprocessing_train/ssd_random_sample_patch_wrapper/cond/Merge_1:0", shape=(?,), dtype=int64)
gbboxes_raw: Tensor("ssd_preprocessing_train/random_flip_left_right/cond_1/Merge:0", shape=(?, 4), dtype=float32)
"""
return save_image_op
def _parse_function(self, example_proto):
"""Function to parse the example proto.
Args:
example_proto: Proto in the format of tf.Example.
Returns:
A dictionary with parsed image, label, height, width and image name.
Raises:
ValueError: Label is of wrong shape.
"""
# Currently only supports jpeg and png.
# Need to use this logic because the shape is not known for
# tf.image.decode_image and we rely on this info to
# extend label if necessary.
def _decode_image(content, channels):
return tf.cond(tf.image.is_jpeg(content),
lambda: tf.image.decode_jpeg(content, channels),
lambda: tf.image.decode_png(content, channels))
features = {
'image/encoded':
tf.FixedLenFeature((), tf.string, default_value=''),
'image/filename':
tf.FixedLenFeature((), tf.string, default_value=''),
'image/format':
tf.FixedLenFeature((), tf.string, default_value='jpeg'),
'image/height':
tf.FixedLenFeature((), tf.int64, default_value=0),
'image/width':
tf.FixedLenFeature((), tf.int64, default_value=0),
'image/segmentation/class/encoded':
tf.FixedLenFeature((), tf.string, default_value=''),
'image/segmentation/class/format':
tf.FixedLenFeature((), tf.string, default_value='png'),
}
parsed_features = tf.parse_single_example(example_proto, features)
image = _decode_image(parsed_features['image/encoded'], channels=3)
label = None
if self.split_name != common.TEST_SET:
label = _decode_image(
parsed_features['image/segmentation/class/encoded'],
channels=1)
image_name = parsed_features['image/filename']
if image_name is None:
image_name = tf.constant('')
sample = {
common.IMAGE: image,
common.IMAGE_NAME: image_name,
common.HEIGHT: parsed_features['image/height'],
common.WIDTH: parsed_features['image/width'],
}
if label is not None:
if label.get_shape().ndims == 2:
label = tf.expand_dims(label, 2)
elif label.get_shape().ndims == 3 and label.shape.dims[2] == 1:
pass
else:
raise ValueError(
'Input label shape must be [height, width], or '
'[height, width, 1].')
label.set_shape([None, None, 1])
sample[common.LABELS_CLASS] = label
return sample
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
name_to_features = {
"input_ids": tf.FixedLenFeature([sea_length], tf.int64),
"input_mask": tf.FixedLenFeature([seq_length], tf.int64),
"aux_data": tf.FixedLenFeature([30], tf.int64),
"segment_ids": tf.FixedLenFeature([seq_length], tf.int64),
"label_ids": tf.FixedLenFeature([], tf.int64),
"is_real_example": tf.FixedLenFeature([], tf.int64),
}
#again using 30 directly in above, although would be better to input in flags
def _decode_record(record, name_to_features):
example = tf.parse_single_example(record, name_to_features)
# tf.Example only supports tf.int64, but the TPU only supports tf.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == tf.int64:
t = tf.to_int32(t)
example[name] = t
return example
def input_fn(params):
batch_size = params["batch_size"]
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
d = None
if is_training:
gap_file = os.path.join(input_file, "*.tsv.tf_record")
gap = tf.data.Dataset.list_files(gap_file)\
.flat_map(tf.data.TFRecordDataset)\
.take(64)
if len(tf.gfile.ListDirectory(input_file)) > 3:
pseudo_file = os.path.join(input_file, "*-tsv.tf_record")
pseudo = tf.data.Dataset.list_files(pseudo_file)\
.flat_map(tf.data.TFRecordDataset)\
.shuffle(buffer_size=1000)\
.take(128)
d = gap.concatenate(pseudo).shuffle(buffer_size=128)
else:
d = tf.data.Dataset.list_files(gap_file)\
.flat_map(tf.data.TFRecordDataset)
d = d.repeat()
else:
d = tf.data.TFRecordDataset(input_file)
d = d.apply(
tf.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
path = "/media/hlee/Work/set/output.tfrecords"
files = tf.train.match_filenames_once(path) #获取所有符合正则表达式的文件,返回文件列表
filename_queue = tf.train.string_input_producer(
[path], shuffle=False) # create a queue
reader = tf.TFRecordReader()
_, serialized_example = reader.read(
filename_queue) # return file_name and file
features = tf.parse_single_example(serialized_example,
features={
'label': tf.FixedLenFeature([],
tf.int64),
'img': tf.FixedLenFeature([],
tf.string),
'width': tf.FixedLenFeature([],
tf.int64),
'height': tf.FixedLenFeature([],
tf.int64),
'channels': tf.FixedLenFeature([],
tf.int64)
}) # return image and label
# img = test_tf.image.convert_image_dtype(img, dtype=test_tf.float32)
# img = test_tf.reshape(img, [512, 80, 3]) # reshape image to 512*80*3
# img = test_tf.cast(img, test_tf.float32) * (1. / 255) - 0.5 # throw img tensor
label = tf.cast(features['label'], tf.int32) # throw label tensor
def getSplit(self, split_name):
###############################################################
#
# Obtains the training/validation split
#
###############################################################
#Check whether the split_name is train or validation
if split_name not in ['train', 'validation']:
raise ValueError(
'The split_name %s is not recognized. Please input either train or validation as the split_name'
% (split_name))
#Create the full path for a general FilePattern to locate the tfrecord_files
FilePattern_path = os.path.join(
self.confs["Classifier"]["DatasetDir"],
self.confs["Classifier"]["FilePattern"] % (split_name))
#Count the total number of examples in all of these shard
num_samples = 0
FilePattern_for_counting = '200label_' + split_name
tfrecords_to_count = [
os.path.join(self.confs["Classifier"]["DatasetDir"], file)
for file in os.listdir(self.confs["Classifier"]["DatasetDir"])
if file.startswith(FilePattern_for_counting)
]
#print(tfrecords_to_count)
for tfrecord_file in tfrecords_to_count:
for record in tf.python_io.tf_record_iterator(tfrecord_file):
num_samples += 1
#Create a reader, which must be a TFRecord reader in this case
reader = tf.TFRecordReader
#Create the keys_to_features dictionary for the decoder
keys_to_features = {
'image/encoded':
tf.FixedLenFeature((), tf.string, default_value=''),
'image/format':
tf.FixedLenFeature((), tf.string, default_value='jpg'),
'image/class/label':
tf.FixedLenFeature([],
tf.int64,
default_value=tf.zeros([], dtype=tf.int64)),
}
#Create the items_to_handlers dictionary for the decoder.
items_to_handlers = {
'image': slim.tfexample_decoder.Image(),
'label': slim.tfexample_decoder.Tensor('image/class/label'),
}
#Start to create the decoder
decoder = slim.tfexample_decoder.TFExampleDecoder(
keys_to_features, items_to_handlers)
#Create the labels_to_name file
labels_to_name_dict = self.labelsToName
#Actually create the dataset
dataset = slim.dataset.Dataset(
data_sources=FilePattern_path,
decoder=decoder,
reader=reader,
num_readers=4,
num_samples=num_samples,
num_classes=self.confs["Classifier"]["NumClasses"],
labels_to_name=labels_to_name_dict,
items_to_descriptions=self.items_to_descriptions)
return dataset
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
name_to_features = {
"input_ids": tf.FixedLenFeature([seq_length], tf.int64),
"input_mask": tf.FixedLenFeature([seq_length], tf.float32),
"segment_ids": tf.FixedLenFeature([seq_length], tf.int64),
"label_ids": tf.FixedLenFeature([], tf.int64),
"is_real_example": tf.FixedLenFeature([], tf.int64),
}
if FLAGS.is_regression:
name_to_features["label_ids"] = tf.FixedLenFeature([], tf.float32)
tf.logging.info("Input tfrecord file {}".format(input_file))
def _decode_record(record, name_to_features):
"""Decodes a record to a TensorFlow example."""
example = tf.parse_single_example(record, name_to_features)
# tf.Example only supports tf.int64, but the TPU only supports tf.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == tf.int64:
t = tf.cast(t, tf.int32)
example[name] = t
return example
def input_fn(params, input_context=None):
"""The actual input function."""
if FLAGS.use_tpu:
batch_size = params["batch_size"]
elif is_training:
batch_size = FLAGS.train_batch_size
elif FLAGS.do_eval:
batch_size = FLAGS.eval_batch_size
else:
batch_size = FLAGS.predict_batch_size
d = tf.data.TFRecordDataset(input_file)
# Shard the dataset to difference devices
if input_context is not None:
tf.logging.info("Input pipeline id %d out of %d",
input_context.input_pipeline_id,
input_context.num_replicas_in_sync)
d = d.shard(input_context.num_input_pipelines,
input_context.input_pipeline_id)
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
if is_training:
d = d.shuffle(buffer_size=FLAGS.shuffle_buffer)
d = d.repeat()
d = d.apply(
tf.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
def get_split(split_name, dataset_dir, file_pattern, reader, split_to_sizes,
items_to_descriptions, num_classes):
"""Gets a dataset tuple with instructions for reading Pascal VOC dataset.
Args:
split_name: A train/test split name.
dataset_dir: The base directory of the dataset sources.
file_pattern: The file pattern to use when matching the dataset sources.
It is assumed that the pattern contains a '%s' string so that the split
name can be inserted.
reader: The TensorFlow reader type.
Returns:
A `Dataset` namedtuple.
Raises:
ValueError: if `split_name` is not a valid train/test split.
"""
if split_name not in split_to_sizes:
raise ValueError('split name %s was not recognized.' % split_name)
file_pattern = os.path.join(dataset_dir, file_pattern % split_name)
# Allowing None in the signature so that dataset_factory can use the default.
if reader is None:
reader = tf.TFRecordReader
# Features in Pascal VOC TFRecords.
keys_to_features = {
'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
'image/format': tf.FixedLenFeature((), tf.string,
default_value='jpeg'),
'image/height': tf.FixedLenFeature([1], tf.int64),
'image/width': tf.FixedLenFeature([1], tf.int64),
'image/channels': tf.FixedLenFeature([1], tf.int64),
'image/shape': tf.FixedLenFeature([3], tf.int64),
'image/object/bbox/xmin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/xmax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/label': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/difficult': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/truncated': tf.VarLenFeature(dtype=tf.int64),
}
items_to_handlers = {
'image':
slim.tfexample_decoder.Image('image/encoded', 'image/format'),
'shape':
slim.tfexample_decoder.Tensor('image/shape'),
'object/bbox':
slim.tfexample_decoder.BoundingBox(['ymin', 'xmin', 'ymax', 'xmax'],
'image/object/bbox/'),
'object/label':
slim.tfexample_decoder.Tensor('image/object/bbox/label'),
'object/difficult':
slim.tfexample_decoder.Tensor('image/object/bbox/difficult'),
'object/truncated':
slim.tfexample_decoder.Tensor('image/object/bbox/truncated'),
}
decoder = slim.tfexample_decoder.TFExampleDecoder(keys_to_features,
items_to_handlers)
labels_to_names = None
if dataset_utils.has_labels(dataset_dir):
labels_to_names = dataset_utils.read_label_file(dataset_dir)
# else:
# labels_to_names = create_readable_names_for_imagenet_labels()
# dataset_utils.write_label_file(labels_to_names, dataset_dir)
return slim.dataset.Dataset(data_sources=file_pattern,
reader=reader,
decoder=decoder,
num_samples=split_to_sizes[split_name],
items_to_descriptions=items_to_descriptions,
num_classes=num_classes,
labels_to_names=labels_to_names)
def style_image_inputs(style_dataset_file,
batch_size=None,
image_size=None,
square_crop=False,
shuffle=True):
"""Loads a batch of random style image given the path of tfrecord dataset.
Args:
style_dataset_file: str, path to the tfrecord dataset of style files.
The dataset is produced via the create_style_dataset.py script and is
made of Example protobufs with the following features:
* 'image_raw': byte encoding of the JPEG string of the style image.
* 'label': integer identifier of the style image in [0, N - 1], where
N is the number of examples in the dataset.
* 'vgg_16/<LAYER_NAME>': Gram matrix at layer <LAYER_NAME> of the VGG-16
network (<LAYER_NAME> in {conv,pool}{1,2,3,4,5}) for the style
image.
batch_size: int. If provided, batches style images. Defaults to None.
image_size: int. The images will be resized bilinearly so that the smallest
side has size image_size. Defaults to None.
square_crop: bool. If True, square-crops to [image_size, image_size].
Defaults to False.
shuffle: bool, whether to shuffle style files at random. Defaults to True.
Returns:
If batch_size is defined, a 4-D tensor of shape [batch_size, ?, ?, 3] with
values in [0, 1] for the style image, and 1-D tensor for the style label.
Raises:
ValueError: if center cropping is requested but no image size is provided,
or if batch size is specified but center-cropping is not requested.
"""
vgg_layers = [
'vgg_16/conv1', 'vgg_16/pool1', 'vgg_16/conv2', 'vgg_16/pool2',
'vgg_16/conv3', 'vgg_16/pool3', 'vgg_16/conv4', 'vgg_16/pool4',
'vgg_16/conv5', 'vgg_16/pool5'
]
if square_crop and image_size is None:
raise ValueError('center-cropping requires specifying the image size.')
if batch_size is not None and not square_crop:
raise ValueError('batching requires center-cropping.')
with tf.name_scope('style_image_processing'):
filename_queue = tf.train.string_input_producer([style_dataset_file],
shuffle=False,
capacity=1,
name='filename_queue')
if shuffle:
examples_queue = tf.RandomShuffleQueue(
capacity=64,
min_after_dequeue=32,
dtypes=[tf.string],
name='random_examples_queue')
else:
examples_queue = tf.FIFOQueue(capacity=64,
dtypes=[tf.string],
name='fifo_examples_queue')
reader = tf.TFRecordReader()
_, value = reader.read(filename_queue)
enqueue_ops = [examples_queue.enqueue([value])]
tf.train.queue_runner.add_queue_runner(
tf.train.queue_runner.QueueRunner(examples_queue, enqueue_ops))
example_serialized = examples_queue.dequeue()
features = tf.parse_single_example(
example_serialized,
features={
'label': tf.FixedLenFeature([], tf.int64),
'image_raw': tf.FixedLenFeature([], tf.string),
'vgg_16/conv1': tf.FixedLenFeature([64, 64], tf.float32),
'vgg_16/pool1': tf.FixedLenFeature([64, 64], tf.float32),
'vgg_16/conv2': tf.FixedLenFeature([128, 128], tf.float32),
'vgg_16/pool2': tf.FixedLenFeature([128, 128], tf.float32),
'vgg_16/conv3': tf.FixedLenFeature([256, 256], tf.float32),
'vgg_16/pool3': tf.FixedLenFeature([256, 256], tf.float32),
'vgg_16/conv4': tf.FixedLenFeature([512, 512], tf.float32),
'vgg_16/pool4': tf.FixedLenFeature([512, 512], tf.float32),
'vgg_16/conv5': tf.FixedLenFeature([512, 512], tf.float32),
'vgg_16/pool5': tf.FixedLenFeature([512, 512], tf.float32)
})
image = tf.image.decode_jpeg(features['image_raw'])
label = features['label']
gram_matrices = [features[vgg_layer] for vgg_layer in vgg_layers]
image.set_shape([None, None, 3])
if image_size:
if square_crop:
image = _aspect_preserving_resize(image, image_size + 2)
image = _central_crop([image], image_size, image_size)[0]
image.set_shape([image_size, image_size, 3])
else:
image = _aspect_preserving_resize(image, image_size)
image = tf.to_float(image) / 255.0
if batch_size is None:
image = tf.expand_dims(image, 0)
else:
image_label_gram_matrices = tf.train.batch([image, label] +
gram_matrices,
batch_size=batch_size)
image, label = image_label_gram_matrices[:2]
gram_matrices = image_label_gram_matrices[2:]
gram_matrices = dict([
(vgg_layer, gram_matrix)
for vgg_layer, gram_matrix in zip(vgg_layers, gram_matrices)
])
return image, label, gram_matrices
def get_split(split_name, dataset_dir, file_pattern=None, reader=None):
"""Gets a dataset tuple with instructions for reading ImageNet.
Args:
split_name: A train/test split name.
dataset_dir: The base directory of the dataset sources.
file_pattern: The file pattern to use when matching the dataset sources.
It is assumed that the pattern contains a '%s' string so that the split
name can be inserted.
reader: The TensorFlow reader type.
Returns:
A `Dataset` namedtuple.
Raises:
ValueError: if `split_name` is not a valid train/test split.
"""
if split_name not in _SPLITS_TO_SIZES:
raise ValueError('split name %s was not recognized.' % split_name)
if not file_pattern:
file_pattern = _FILE_PATTERN
file_pattern = os.path.join(dataset_dir, file_pattern % split_name)
# Allowing None in the signature so that dataset_factory can use the default.
if reader is None:
reader = tf.TFRecordReader
keys_to_features = {
'image/encoded':
tf.FixedLenFeature((), tf.string, default_value=''),
'image/format':
tf.FixedLenFeature((), tf.string, default_value='jpeg'),
'image/class/label':
tf.FixedLenFeature([], dtype=tf.int64, default_value=-1),
'image/class/text':
tf.FixedLenFeature([], dtype=tf.string, default_value=''),
'image/object/bbox/xmin':
tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymin':
tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/xmax':
tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymax':
tf.VarLenFeature(dtype=tf.float32),
'image/object/class/label':
tf.VarLenFeature(dtype=tf.int64),
}
items_to_handlers = {
'image':
slim.tfexample_decoder.Image('image/encoded', 'image/format'),
'label':
slim.tfexample_decoder.Tensor('image/class/label'),
'label_text':
slim.tfexample_decoder.Tensor('image/class/text'),
'object/bbox':
slim.tfexample_decoder.BoundingBox(['ymin', 'xmin', 'ymax', 'xmax'],
'image/object/bbox/'),
'object/label':
slim.tfexample_decoder.Tensor('image/object/class/label'),
}
decoder = slim.tfexample_decoder.TFExampleDecoder(keys_to_features,
items_to_handlers)
labels_to_names = None
if LOAD_READABLE_NAMES:
if dataset_utils.has_labels(dataset_dir):
labels_to_names = dataset_utils.read_label_file(dataset_dir)
else:
labels_to_names = create_readable_names_for_imagenet_labels()
dataset_utils.write_label_file(labels_to_names, dataset_dir)
return slim.dataset.Dataset(data_sources=file_pattern,
reader=reader,
decoder=decoder,
num_samples=_SPLITS_TO_SIZES[split_name],
items_to_descriptions=_ITEMS_TO_DESCRIPTIONS,
num_classes=_NUM_CLASSES,
labels_to_names=labels_to_names)
import cv2
MOVING_AVERAGE_DECAY = 0.99
EVAL_INTERVAL_SECS = 10
image_size = 128
batch_size = 200
files = tf.train.match_filenames_once("train.tfrecords")
filename_queue = tf.train.string_input_producer(files, shuffle=True)
# filename_queue = tf.train.string_input_producer(["dog_train.tfrecords0"]) #读入流中
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue) # 返回文件名和文件
features = tf.parse_single_example(serialized_example,
features={
'height':
tf.FixedLenFeature([], tf.int64),
'width':
tf.FixedLenFeature([], tf.int64),
'label':
tf.FixedLenFeature([], tf.int64),
'img_raw':
tf.FixedLenFeature([], tf.string),
'channels':
tf.FixedLenFeature([], tf.int64),
}) # 取出包含image和label的feature对象
image, label = features['img_raw'], features['label']
height, width = features['height'], features['width']
height = tf.cast(height, tf.uint8)
width = tf.cast(width, tf.float32)
channels = features['channels']
decoded_image = tf.decode_raw(image, tf.uint8)
import tensorflow_transform as tft
from string import ascii_lowercase
from apache_beam.io import tfrecordio
from tensorflow_transform import coders
from tensorflow_transform.beam import impl as beam_impl
from tensorflow_transform.beam.tft_beam_io import transform_fn_io
from tensorflow_transform.tf_metadata import dataset_metadata, metadata_io,dataset_schema
DELIMITERS_WORDS=' '
MAX_WORD_LEN=15
MAX_SENTENCE_LEN=25
# IT IS EXTREMELY COSTFUL TO HAVE TO LONG WORDS/SENTENCES
TRAIN_DATA_SCHEMA = dataset_schema.from_feature_spec({
'id': tf.FixedLenFeature(shape=[], dtype=tf.float32),
'text': tf.FixedLenFeature(shape=[], dtype=tf.string),
'labels': tf.FixedLenFeature(shape=[MAX_SENTENCE_LEN], dtype=tf.int64),
'label': tf.FixedLenFeature(shape=[], dtype=tf.string),
"chars" : tf.FixedLenFeature(shape=[MAX_SENTENCE_LEN,MAX_WORD_LEN], dtype=tf.int64),
'label_length': tf.FixedLenFeature(shape=[], dtype=tf.int64),
'sentence_length': tf.FixedLenFeature(shape=[], dtype=tf.int64),
'chars_in_word': tf.FixedLenFeature(shape=[MAX_SENTENCE_LEN], dtype=tf.int64),
})
train_metadata = dataset_metadata.DatasetMetadata(schema=TRAIN_DATA_SCHEMA)
OUTPUT_FILE_TRAIN="train_records.tfrecords"
MAPPING = {a:index for index,a in enumerate(ascii_lowercase + ascii_lowercase.upper())}
UNKONWN = len(MAPPING)
PADD_VALUE_CHAR = UNKONWN+1
def main(_):
if len(sys.argv) < 2 or sys.argv[-1].startswith('-'):
print(
'Usage: mnist_export.py [--training_iteration=x] '
'[--model_version=y] export_dir')
sys.exit(-1)
if FLAGS.training_iteration <= 0:
print 'Please specify a positive value for training iteration.'
sys.exit(-1)
if FLAGS.model_version <= 0:
print 'Please specify a positive value for version number.'
sys.exit(-1)
# Train model
print 'Training model...'
mnist = mnist_input_data.read_data_sets(FLAGS.work_dir, one_hot=True)
sess = tf.InteractiveSession()
serialized_tf_example = tf.placeholder(tf.string, name='tf_example')
feature_configs = {
'x': tf.FixedLenFeature(shape=[784], dtype=tf.float32),
}
tf_example = tf.parse_example(serialized_tf_example, feature_configs)
x = tf.identity(tf_example['x'],
name='x') # use tf.identity() to assign name
y_ = tf.placeholder('float', shape=[None, 10])
w = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
sess.run(tf.global_variables_initializer())
y = tf.nn.softmax(tf.matmul(x, w) + b, name='y')
cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(
cross_entropy)
values, indices = tf.nn.top_k(y, 10)
table = tf.contrib.lookup.index_to_string_table_from_tensor(
tf.constant([str(i) for i in xrange(10)]))
prediction_classes = table.lookup(tf.to_int64(indices))
for _ in range(FLAGS.training_iteration):
batch = mnist.train.next_batch(50)
train_step.run(feed_dict={x: batch[0], y_: batch[1]})
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, 'float'))
print 'training accuracy %g' % sess.run(accuracy,
feed_dict={
x: mnist.test.images,
y_: mnist.test.labels
})
print 'Done training!'
# Export model
# WARNING(break-tutorial-inline-code): The following code snippet is
# in-lined in tutorials, please update tutorial documents accordingly
# whenever code changes.
export_path_base = sys.argv[-1]
export_path = os.path.join(tf.compat.as_bytes(export_path_base),
tf.compat.as_bytes(str(FLAGS.model_version)))
print 'Exporting trained model to', export_path
builder = tf.saved_model.builder.SavedModelBuilder(export_path)
# Build the signature_def_map.
classification_inputs = tf.saved_model.utils.build_tensor_info(
serialized_tf_example)
classification_outputs_classes = tf.saved_model.utils.build_tensor_info(
prediction_classes)
classification_outputs_scores = tf.saved_model.utils.build_tensor_info(
values)
classification_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs={
tf.saved_model.signature_constants.CLASSIFY_INPUTS:
classification_inputs
},
outputs={
tf.saved_model.signature_constants.CLASSIFY_OUTPUT_CLASSES:
classification_outputs_classes,
tf.saved_model.signature_constants.CLASSIFY_OUTPUT_SCORES:
classification_outputs_scores
},
method_name=tf.saved_model.signature_constants.CLASSIFY_METHOD_NAME
))
tensor_info_x = tf.saved_model.utils.build_tensor_info(x)
tensor_info_y = tf.saved_model.utils.build_tensor_info(y)
prediction_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs={'images': tensor_info_x},
outputs={'scores': tensor_info_y},
method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME)
)
legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op')
builder.add_meta_graph_and_variables(
sess, [tf.saved_model.tag_constants.SERVING],
signature_def_map={
'predict_images':
prediction_signature,
tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
classification_signature,
},
legacy_init_op=legacy_init_op)
builder.save()
print 'Done exporting!'
image_width = 32
image_channel = 3
class_numbers = 10
#model storage
model_checkpoint_path = join( checkpoint_path, 'model.ckpt')
train_summary_path = join( checkpoint_path, 'train' )
valid_summary_path = join( checkpoint_path, 'valid' )
## data loading
root_path = '/home/rachakra/few_shot_learning/prepare_data/output'
train_tfrecords = join( root_path, 'train_siamese_pair_{0}.tfrecords'.format( sample_per_class ) )
valid_tfrecords = join( root_path, 'siamese_pair_valid.tfrecords' )
## information for parsing the tfrecord
features={'image_one':tf.FixedLenFeature([], tf.string),
'image_two':tf.FixedLenFeature([], tf.string),\
'label': tf.FixedLenFeature([], tf.int64 )}
train_parser = Parser( features, image_height, image_width, True )
valid_parser = Parser( features, image_height, image_width, False )
##test files
training_list_file = 'train_raw_list_{0}.txt'.format( sample_per_class )
class_valid_list = join( root_path, 'class_valid.pickle' )
def create_sprite_image(examples):
"""Returns an encoded sprite image for use in Facets Dive.
Args:
examples: A list of serialized example protos to get images for.
Returns:
An encoded PNG.
"""
def generate_image_from_thubnails(thumbnails, thumbnail_dims):
"""Generates a sprite atlas image from a set of thumbnails."""
num_thumbnails = tf.shape(thumbnails)[0].eval()
images_per_row = int(math.ceil(math.sqrt(num_thumbnails)))
thumb_height = thumbnail_dims[0]
thumb_width = thumbnail_dims[1]
master_height = images_per_row * thumb_height
master_width = images_per_row * thumb_width
num_channels = 3
master = np.zeros([master_height, master_width, num_channels])
for idx, image in enumerate(thumbnails.eval()):
left_idx = idx % images_per_row
top_idx = int(math.floor(idx / images_per_row))
left_start = left_idx * thumb_width
left_end = left_start + thumb_width
top_start = top_idx * thumb_height
top_end = top_start + thumb_height
master[top_start:top_end, left_start:left_end, :] = image
return tf.image.encode_png(master)
image_feature_name = 'image/encoded'
sprite_thumbnail_dim_px = 32
with tf.compat.v1.Session():
keys_to_features = {
image_feature_name:
tf.FixedLenFeature((), tf.string, default_value=''),
}
parsed = tf.parse_example(examples, keys_to_features)
images = tf.zeros([1, 1, 1, 1], tf.float32)
i = tf.constant(0)
thumbnail_dims = (sprite_thumbnail_dim_px, sprite_thumbnail_dim_px)
num_examples = tf.constant(len(examples))
encoded_images = parsed[image_feature_name]
# Loop over all examples, decoding the image feature value, resizing
# and appending to a list of all images.
def loop_body(i, encoded_images, images):
encoded_image = encoded_images[i]
image = tf.image.decode_jpeg(encoded_image, channels=3)
resized_image = tf.image.resize(image, thumbnail_dims)
expanded_image = tf.expand_dims(resized_image, 0)
images = tf.cond(tf.equal(i, 0), lambda: expanded_image,
lambda: tf.concat([images, expanded_image], 0))
return i + 1, encoded_images, images
loop_out = tf.while_loop(
lambda i, encoded_images, images: tf.less(i, num_examples),
loop_body, [i, encoded_images, images],
shape_invariants=[
i.get_shape(),
encoded_images.get_shape(),
tf.TensorShape(None)
])
# Create the single sprite atlas image from these thumbnails.
sprite = generate_image_from_thubnails(loop_out[2], thumbnail_dims)
return sprite.eval()
def Train(graph, Batch_size, Iteration):
filenames = [("train_1.tfrecords")]
print(filenames)
filenames_cross = ["train_2.tfrecords"]
filename_queue = tf.train.string_input_producer(filenames)
# Tensorflow需要队列流输入数据
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
# tf.parse_single_example为解析器
features = tf.parse_single_example(serialized_example,
features={
"Mat":
tf.FixedLenFeature([], tf.string),
"label":
tf.FixedLenFeature([], tf.int64),
})
Mat_32 = tf.decode_raw(features["Mat"], tf.float64)
Mat_32 = tf.reshape(Mat_32, [32, 32, 1])
Label_32 = tf.cast(features["label"], tf.float32)
# Mat_batch,Label64_batch,Label32_batch,Label32_Batch = tf.train.batch\
# ([Mat_64, Label_64,Label_32,Label_32],batch_size= Batch_size,capacity=300,num_threads= 1)
# tf.train.shuffle_batch（）这个函数对读取到的数据进行了batch处理，这样更有利于后续的训练。
Mat_batch, Label32_batch = tf.train.shuffle_batch \
([Mat_32, Label_32], batch_size=Batch_size, capacity=3 * Batch_size, min_after_dequeue=5,
num_threads=32)
print("\n--------- finish reading train data-----------")
filename_queue_cross = tf.train.string_input_producer(filenames_cross)
reader_cross = tf.TFRecordReader()
_, serialized_example_cross = reader_cross.read(filename_queue_cross)
features_cross = tf.parse_single_example(serialized_example_cross,
features={
"Mat":
tf.FixedLenFeature([],
tf.string),
"label":
tf.FixedLenFeature([],
tf.int64),
})
Mat_32_cross = tf.decode_raw(features_cross["Mat"], tf.float64)
Mat_32_cross = tf.reshape(Mat_32_cross, [32, 32, 1])
Label_32_cross = tf.cast(features_cross["label"], tf.float32)
Mat_batch_cross, Label32_batch_cross = tf.train.shuffle_batch \
([Mat_32_cross, Label_32_cross], batch_size=Batch_size,
capacity=Batch_size * 3, min_after_dequeue=5,
num_threads=32)
# saver = tf.train.Saver()
# print("Mat_Batttt",Mat_Batch)
with tf.Session() as sess:
# 开启协调器
coord = tf.train.Coordinator()
# 使用start_queue_runners 启动队列填充
# 一定要开启队列，要不然是空的
threads = tf.train.start_queue_runners(sess, coord)
initglo = tf.global_variables_initializer()
initloc = tf.local_variables_initializer()
sess.run(initglo)
sess.run(initloc)
tf.local_variables_initializer().run()
epoch = 9
learning_rate = 0.01
for j in range(epoch):
if j % 3 == 0:
learning_rate = learning_rate * 0.1
print(learning_rate)
acc = 0
loss = 0
acc_cross = 0
loss_cross = 0
for i in range(Iteration):
# 这句话的意思是从队列里拿出指定批次的数据
is_training = True
mat_32, label_32 = sess.run([Mat_batch, Label32_batch])
mat_32 = mat_32.reshape(Batch_size, 32, 32, 1)
# print("Mat_64 is ",Mat_64)
# print("Label_64 is ",Label_64)
mat_32 = mat_32.astype('float32')
label_32 = label_32.astype("float32")
label_32 = label_32.reshape(Batch_size, 1)
for b in range(Batch_size):
mat_32[b] = LoadData.Normalize(mat_32[b])
# LOSS_64, acc_rate64, Pre_lable_64, _ = sess.run(
# [graph["LOSS"], graph["acc"], graph["Pre"], graph["Train"]],
# feed_dict={
# graph["inputs_Mat"]: mat_64,
# graph["ys_label"]: label_64,
# graph['is_training']: is_training,
# graph["is_64"]: 1,
# graph["is_32"]: 0,
# graph["kp"]: 0.5,
# graph["ln"]:learning_rate,
# })
# LOSS_32, acc_rate32, Pre_lable_32, _ = sess.run(
# [graph["LOSS"], graph["acc"], graph["Pre"], graph["Train"]],
# feed_dict={
# graph["inputs_Mat"]: mat_32,
# graph["ys_label"]: label_32,
# graph['is_training']: is_training,
# graph["is_64"]: 0,
# graph["is_32"]: 1,
# graph["kp"]: 0.5,
# graph["ln"]: learning_rate,
# })
# print(label_32)
LOSS_32, acc_rate32, Pre_lable_32, _ = sess.run(
[
graph["LOSS"], graph["acc"], graph["Pre"],
graph["Train"]
],
feed_dict={
graph["inputs_Mat"]: mat_32,
graph["ys_label"]: label_32,
graph['is_training']: is_training,
graph["kp"]: 0.5,
graph["ln"]: learning_rate,
})
if i % 100 == 0:
is_training = False
print("\n------------- Iteration %d --------------" % (i))
# # print("LOSS64 is ", LOSS_64)
# # print("LOSS32 is ", LOSS_32)
# print("LOSS16 is ", LOSS_16)
# # print("acc_rate64 is : ", acc_rate64)
# # print("acc_rate32 is : ", acc_rate32)
# print("acc_rate16 is : ", acc_rate16)
# # 把输出写入到文件里
# stdout_backup = sys.stdout
# log_file = open(YUV + "_RunDetail.log", "a")
# sys.stdout = log_file
# print("--------------- Iteration %d ---------------" % (i))
#
# print("LOSS64 is ", LOSS_64)
# print("LOSS32 is ", LOSS_32)
# print("LOSS16 is ", LOSS_16)
# print("acc_rate64 is : ", acc_rate64)
# print("acc_rate32 is : ", acc_rate32)
# print("acc_rate16 is : ", acc_rate16)
mat_cross32, label_cross32 = sess.run(
[Mat_batch_cross, Label32_batch_cross])
mat_cross32 = mat_cross32.reshape(Batch_size, 32, 32, 1)
mat_cross32 = mat_cross32.astype("float32")
for b in range(Batch_size):
mat_cross32[b] = LoadData.Normalize(mat_cross32[b])
label_cross32 = label_cross32.astype("float32")
label_cross32 = label_cross32.reshape(Batch_size, 1)
LOSS_ceoss32, acc_rate_cross32, Pre_lable_cross32 = sess.run(
[graph["LOSS"], graph["acc"], graph["Pre"]],
feed_dict={
graph["inputs_Mat"]: mat_cross32,
graph["ys_label"]: label_cross32,
graph['is_training']: is_training,
graph["kp"]: 1.0,
})
acc_cross = acc_cross + acc_rate_cross32
loss_cross = loss_cross + LOSS_ceoss32
loss = loss + LOSS_32
acc = acc + acc_rate32
print("*************train*************")
print("acc:" + str(acc / 60))
print("loss:" + str(loss / 60))
print("*************cross*************")
print("acc:" + str(acc_cross / 60))
print("loss:" + str(loss_cross / 60))
# sum16 = 0
# a16 = 0
# for e in range(Batch_size):
# if label_cross16[e][0] == 1:
# sum16 += 1
# if Pre_lable_cross16[e][0] == 1:
# a16 += 1
# if sum16 != 0:
# print("reCall 16 is : ", a16 / sum16)
# sum16 = 0
# a16 = 0
# for e in range(Batch_size):
# if label_cross16[e][0] == 0:
# sum16 += 1
# if Pre_lable_cross16[e][0] == 0:
# a16 += 1
# if sum16 != 0:
# print("Call 16 is : ", a16 / sum16)
if j == 8:
print("save the model finally:" + str(i))
var_list = [
var for var in tf.global_variables()
if "moving" in var.name
]
var_list += tf.trainable_variables()
saver = tf.train.Saver(var_list=var_list, max_to_keep=20)
Write_List = ['X', 'is_train', 'kp', 'y', 'acc', 'output']
constant_graph = graph_util.convert_variables_to_constants \
(sess, sess.graph_def, Write_List)
with tf.gfile.FastGFile(pb_file_path, mode='wb') as f:
f.write(constant_graph.SerializeToString())
saver.save(sess, "model/Demo-model")
# saver.save(sess, "./model/Demo-model-one")
print("save the cross model")
coord.request_stop()
print("program ending")
coord.join(threads)