Related
I'm trying to use keras ImageDataGenerator for training a pix2pix CNN model. It maps input images to output images. We know that the keras ImageDataGenerator can be used easily for image classification, but I'm having problems to train a pix2pix model. Here is my attempt:
Custom generator:
class JoinedGen(tf.keras.utils.Sequence):
def __init__(self, input_gen, target_gen):
self.input_gen = input_gen
self.target_gen = target_gen
assert len(input_gen) == len(target_gen)
def __len__(self):
return len(self.input_gen)
def __getitem__(self, i):
x = self.input_gen[i]
y = self.target_gen[i]
return x, y
def on_epoch_end(self):
self.input_gen.on_epoch_end()
self.target_gen.on_epoch_end()
self.target_gen.index_array = self.input_gen.index_array
Implementation with ImageDataGenerator:
generator = ImageDataGenerator(shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
validation_split=0.3)
input_gen = generator.flow_from_directory(path,
classes=['area'],
shuffle=False,
target_size=(256, 256),
class_mode=None,
batch_size=32,
subset='training')
target_gen = generator.flow_from_directory(path,
classes=['sat'],
shuffle=False,
target_size=(256, 256),
class_mode=None,
batch_size=32,
subset='training')
input_gen_val = generator.flow_from_directory(path,
classes=['area'],
shuffle=False,
target_size=(256, 256),
class_mode=None,
batch_size=32,
subset='validation')
target_gen_val = generator.flow_from_directory(path,
classes=['sat'],
shuffle=False,
target_size=(256, 256),
class_mode=None,
batch_size=32,
subset='validation')
But when I ask for the first image of both training generators using input_gen.next()[0] and target_gen.next()[0] it doesn't give me the corresponding input and output!
As it is said in the Keras documentation the solution is to "provide the same seed and keyword arguments to the fit and flow methods - seed = 1".
Just add to the flow_from_directory method seed = 1.
Check out the link for more information here
I have to build a neural network that can recognize the face of 15 people. I'm using keras. My dataset is composed of 300 total images and is divided into Training, Validation and Test. For each of the 15 people I have the following subdivision:
Training: 13
Validation: 3
Test: 4
Since I couldn't build an efficient neural network from scratch, I also believe because my dataset is very small, I'm trying to solve my problem by doing transfer learning. I used the vgg16 network. In the training and validation phase I get good results but when I run the tests the results are disastrous.
I don't know what my problem is. Here is the code I used:
img_width, img_height = 256, 256
train_data_dir = 'dataset_biometria/face/training_set'
validation_data_dir = 'dataset_biometria/face/validation_set'
nb_train_samples = 20
nb_validation_samples = 20
batch_size = 16
epochs = 5
model = applications.VGG19(weights = "imagenet", include_top=False, input_shape = (img_width, img_height, 3))
for layer in model.layers:
layer.trainable = False
#Adding custom Layers
x = model.output
x = Flatten()(x)
x = Dense(1024, activation="relu")(x)
x = Dropout(0.4)(x)
x = Dense(1024, activation="relu")(x)
predictions = Dense(15, activation="softmax")(x)
# creating the final model
model_final = Model(input = model.input, output = predictions)
# compile the model
model_final.compile(loss = "categorical_crossentropy", optimizer = optimizers.SGD(lr=0.0001, momentum=0.9), metrics=["accuracy"])
# Initiate the train and test generators with data Augumentation
train_datagen = ImageDataGenerator(
rescale = 1./255,
horizontal_flip = True,
fill_mode = "nearest",
zoom_range = 0.3,
width_shift_range = 0.3,
height_shift_range=0.3,
rotation_range=30)
test_datagen = ImageDataGenerator(
rescale = 1./255,
horizontal_flip = True,
fill_mode = "nearest",
zoom_range = 0.3,
width_shift_range = 0.3,
height_shift_range=0.3,
rotation_range=30)
train_generator = train_datagen.flow_from_directory(
train_data_dir,
target_size = (img_height, img_width),
batch_size = batch_size,
class_mode = "categorical")
validation_generator = test_datagen.flow_from_directory(
validation_data_dir,
target_size = (img_height, img_width),
class_mode = "categorical")
# Save the model according to the conditions
checkpoint = ModelCheckpoint("vgg16_1.h5", monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=False, mode='auto', period=1)
early = EarlyStopping(monitor='val_acc', min_delta=0, patience=10, verbose=1, mode='auto')
# Train the model
model_final.fit_generator(
train_generator,
samples_per_epoch = nb_train_samples,
epochs = epochs,
validation_data = validation_generator,
nb_val_samples = nb_validation_samples,
callbacks = [checkpoint, early])
model('model_face_classification.h5')
I also tried to train some layers instead of not training any, as in the example below:
for layer in model.layers[:10]:
layer.trainable = False
I also tried changing the number of epochs, batch size, nb_validation_samples, nb_validation_sample.
Unfortunately the result has not changed, in the testing phase my network cannot correctly recognize faces.
Without seeing the actual results or errors I can not say what the problem is here.
Definitely, small dataset is a problem, but there are many ways to get around it.
You can use image augmentation to increase the samples. You can refer augement.py.
But instead of modifying your above network, there is a really cool model : siamese network/one-shot learning. It does not need too many pics and the accuracies are great.
Therefore you can see below links to get some help :
Facial-Recognition-Using-FaceNet-Siamese-One-Shot-Learning
Face-recognition-using-deep-learning
I have a CNN that saves the bottleneck features of the training and test data with the VGG16 architecture, then uploads the features to my custom fully connected layers to classify the images.
#create data augmentations for training set; helps reduce overfitting and find more features
train_datagen = ImageDataGenerator(rescale=1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip=True)
#use ImageDataGenerator to upload validation images; data augmentation not necessary for
validating process
val_datagen = ImageDataGenerator(rescale=1./255)
#load VGG16 model, pretrained on imagenet database
model = applications.VGG16(include_top=False, weights='imagenet')
#generator to load images into NN
train_generator = train_datagen.flow_from_directory(
train_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode=None,
shuffle=False)
#total number of images used for training data
num_train = len(train_generator.filenames)
#save features to numpy array file so features do not overload memory
bottleneck_features_train = model.predict_generator(train_generator, num_train // batch_size)
val_generator = val_datagen.flow_from_directory(
val_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode=None,
shuffle=False)
num_val = len(val_generator.filenames)
bottleneck_features_validation = model.predict_generator(val_generator, num_val // batch_size)`
#used to retrieve the labels of the images
label_datagen = ImageDataGenerator(rescale=1./255)
#generators can create class labels for each image in either
train_label_generator = label_datagen.flow_from_directory(
train_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode=None,
shuffle=False)
#total number of images used for training data
num_train = len(train_label_generator.filenames)
#load features from VGG16 and pair each image with corresponding label (0 for normal, 1 for pneumonia)
#train_data = np.load('xray/bottleneck_features_train.npy')
#get the class labels generated by train_label_generator
train_labels = train_label_generator.classes
val_label_generator = label_datagen.flow_from_directory(
val_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode=None,
shuffle=False)
num_val = len(val_label_generator.filenames)
#val_data = np.load('xray/bottleneck_features_validation.npy')
val_labels = val_label_generator.classes
#create fully connected layers, replacing the ones cut off from the VGG16 model
model = Sequential()
#converts model's expected input dimensions to same shape as bottleneck feature arrays
model.add(Flatten(input_shape=bottleneck_features_train.shape[1:]))
#ignores a fraction of input neurons so they do not become co-dependent on each other; helps prevent overfitting
model.add(Dropout(0.7))
#normal fully-connected layer with relu activation. Replaces all negative inputs with 0 and does not fire neuron,
#creating a lighetr network
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.7))
#output layer to classify 0 or 1
model.add(Dense(1, activation='sigmoid'))
#compile model and specify which optimizer and loss function to use
#optimizer used to update the weights to optimal values; adam optimizer maintains seperate learning rates
#for each weight and updates accordingly
#cross-entropy function measures the ability of model to correctly classify 0 or 1
model.compile(optimizer=optimizers.Adam(lr=0.0007), loss='binary_crossentropy', metrics=['accuracy'])
#used to stop training if NN shows no improvement for 5 epochs
early_stop = EarlyStopping(monitor='val_loss', min_delta=0.01, patience=5, verbose=1)
#checks each epoch as it runs and saves the weight file from the model with the lowest validation loss
checkpointer = ModelCheckpoint(filepath=top_model_weights_dir, verbose=1, save_best_only=True)
#fit the model to the data
history = model.fit(bottleneck_features_train, train_labels,
epochs=epochs,
batch_size=batch_size,
callbacks = [early_stop, checkpointer],
verbose=2,
validation_data=(bottleneck_features_validation, val_labels))`
After calling train_top_model(), the CNN gets an 86% accuracy after around 10 epochs.
However, when I try implementing this architecture in by building the fully connected layers directly on top of the VGG16 layers, The network gets stuck at a val_acc of 0.5000 and basically does not train. Are there any issues with the code?
epochs = 10
batch_size = 20
train_datagen = ImageDataGenerator(rescale=1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip=True)
val_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
train_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='binary',
shuffle=False)
num_train = len(train_generator.filenames)
val_generator = val_datagen.flow_from_directory(
val_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='binary',
shuffle=False)
num_val = len(val_generator.filenames)`
base_model = applications.VGG16(weights='imagenet', include_top=False, input_shape=(img_width,
img_height, 3))
x = base_model.output
x = Flatten()(x)
x = Dropout(0.7)(x)
x = Dense(128, activation='relu')(x)
x = Dropout(0.7)(x)
predictions = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=predictions)
for layer in model.layers[:19]:
layer.trainable = False
checkpointer = ModelCheckpoint(filepath=top_model_weights_dir, verbose=1, save_best_only=True)
model.compile(optimizer=optimizers.Adam(lr=0.0007), loss='binary_crossentropy', metrics=
['accuracy'])
history = model.fit_generator(train_generator,
steps_per_epoch=(num_train//batch_size),
validation_data=val_generator,
validation_steps=(num_val//batch_size),
callbacks=[checkpointer],
verbose=1,
epochs=epochs)
The reason is that in the second approach, you have not frozen the VGG16 layers. In other words, you are training the whole network. Whereas in the first approach you are just training the weights of your fully connected layers.
Use something like this:
for layer in base_model.layers[:end_layer]:
layer.trainable = False
where end_layer is the last layer you are importing.
Beginner to Deep learning..
I'm trying to identify the slum using satellite images(google map) for Pune city. So, in training dataset i have provided about 100 images of slum and 100 images of other area. But my model is not able to classify input image properly even though accuracy rate is high.
I think this might be because of dimensions of image.
I'm resizing all images to 128*128 pixel.
Kernal size is 3*3.
Link to the map:
https://www.google.co.in/maps/#18.5129661,73.822531,286m/data=!3m1!1e3?hl=en
Following is the code
import os,cv2
import glob
import numpy as np
from keras.utils import plot_model
from keras.utils.np_utils import to_categorical
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split
from keras.models import Model
from keras.layers import Input, Convolution2D, MaxPooling2D, Flatten, Dense, Dropout
PATH = os.getcwd()
data_path = PATH + '/dataset/*'
files = glob.glob(data_path)
X = []
for myFiles in files:
image = cv2.imread(myFiles)
image_resize = cv2.resize(image, (256, 256))
X.append(image_resize)
image_data = np.array(X)
image_data = image_data.astype('float32')
image_data /= 255
print("Image_data shape ", image_data.shape)
no_of_classes = 2
no_of_samples = image_data.shape[0]
label = np.ones(no_of_samples, dtype='int64')
label[0:86] = 0 #Slum
label[87:] = 1 #noSlum
Y = to_categorical(label, no_of_classes)
#shuffle dataset
x,y = shuffle(image_data , Y, random_state = 2)
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state = 2)
#print(x_train)
#print(y_train)
input_shape = image_data[0].shape
input = Input(input_shape)
conv_1 = Convolution2D(32,(3,3), padding='same', activation='relu')(input)
conv_2 = Convolution2D(32,(3,3), padding = 'same', activation = 'relu')(conv_1)
pool_1 = MaxPooling2D(pool_size = (2,2))(conv_2)
drop_1 = Dropout(0.5)(pool_1)
conv_3 = Convolution2D(64,(3,3), padding='same', activation='relu')(drop_1)
conv_4 = Convolution2D(64,(3,3), padding='same', activation = 'relu')(conv_3)
pool_2 = MaxPooling2D(pool_size = (2,2))(conv_4)
drop_2 = Dropout(0.5)(pool_2)
flat_1 = Flatten()(drop_2)
hidden = Dense(64,activation='relu')(flat_1)
drop_3 = Dropout(0.5)(hidden)
out = Dense(no_of_classes,activation = 'softmax')(drop_3)
model = Model(inputs = input, outputs = out)
model.compile(loss = 'categorical_crossentropy', optimizer = 'rmsprop', metrics= ['accuracy'])
model.fit(x_train,y_train,batch_size=10,nb_epoch=20,verbose =1, validation_data=(x_test,y_test))
model.save('model.h5')
score = model.evaluate(x_test,y_test,verbose=1)
print('Test Loss: ',score[0])
print('Test Accuracy: ',score[1])
test_image = x_test[0:1]
print(test_image.shape)
print (model.predict(test_image))
Usually, the behavior you've described above resembles to the inability of NN to identify small objects on input images. Just imagine you give an image of 128*128 with rough noise where nothing is seen - you want NN to correctly classify objects?
What to do?
1) Try to manually convert some input image from your dataset to 128*128 size and see on what data you truly train your NN. So, it'll give you more insight --> maybe you need to have better image's dimension size
2) Add more Conv layers with more neurons that will give you ability to detect small and more sophisticated objects by adding more non-linearity to your output function. Google such great Neural Network structures as ResNet.
3) Add more training data, 100 images isn't enough to have an appropriate result
4) Add data augmentation technique as well ( Rotations seem so strong in your case )
And don't give up :) Eventually, you'll solve it out. Good Luck
What is an example of how to use a TensorFlow TFRecord with a Keras Model and tf.session.run() while keeping the dataset in tensors w/ queue runners?
Below is a snippet that works but it needs the following improvements:
Use the Model API
specify an Input()
Load a dataset from a TFRecord
Run through a dataset in parallel (such as with a queuerunner)
Here is the snippet, there are several TODO lines indicating what is needed:
from keras.models import Model
import tensorflow as tf
from keras import backend as K
from keras.layers import Dense, Input
from keras.objectives import categorical_crossentropy
from tensorflow.examples.tutorials.mnist import input_data
sess = tf.Session()
K.set_session(sess)
# Can this be done more efficiently than placeholders w/ TFRecords?
img = tf.placeholder(tf.float32, shape=(None, 784))
labels = tf.placeholder(tf.float32, shape=(None, 10))
# TODO: Use Input()
x = Dense(128, activation='relu')(img)
x = Dense(128, activation='relu')(x)
preds = Dense(10, activation='softmax')(x)
# TODO: Construct model = Model(input=inputs, output=preds)
loss = tf.reduce_mean(categorical_crossentropy(labels, preds))
# TODO: handle TFRecord data, is it the same?
mnist_data = input_data.read_data_sets('MNIST_data', one_hot=True)
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
sess.run(tf.global_variables_initializer())
# TODO remove default, add queuerunner
with sess.as_default():
for i in range(1000):
batch = mnist_data.train.next_batch(50)
train_step.run(feed_dict={img: batch[0],
labels: batch[1]})
print(loss.eval(feed_dict={img: mnist_data.test.images,
labels: mnist_data.test.labels}))
Why is this question relevant?
For high performance training without going back to python
no TFRecord to numpy to tensor conversions
Keras will soon be part of tensorflow
Demonstrate how Keras Model() classes can accept tensors for input data correctly.
Here is some starter information for a semantic segmentation problem example:
example unet Keras model unet.py, happens to be for semantic segmentation.
Keras + Tensorflow Blog Post
An attempt at running the unet model a tf session with TFRecords and a Keras model (not working)
Code to create the TFRecords: tf_records.py
An attempt at running the unet model a tf session with TFRecords and a Keras model is in densenet_fcn.py (not working)
I don't use tfrecord dataset format so won't argue on the pros and cons, but I got interested in extending Keras to support the same.
github.com/indraforyou/keras_tfrecord is the repository. Will briefly explain the main changes.
Dataset creation and loading
data_to_tfrecord and read_and_decode here takes care of creating tfrecord dataset and loading the same. Special care must be to implement the read_and_decode otherwise you will face cryptic errors during training.
Initialization and Keras model
Now both tf.train.shuffle_batch and Keras Input layer returns tensor. But the one returned by tf.train.shuffle_batch don't have metadata needed by Keras internally. As it turns out, any tensor can be easily turned into a tensor with keras metadata by calling Input layer with tensor param.
So this takes care of initialization:
x_train_, y_train_ = ktfr.read_and_decode('train.mnist.tfrecord', one_hot=True, n_class=nb_classes, is_train=True)
x_train_batch, y_train_batch = K.tf.train.shuffle_batch([x_train_, y_train_],
batch_size=batch_size,
capacity=2000,
min_after_dequeue=1000,
num_threads=32) # set the number of threads here
x_train_inp = Input(tensor=x_train_batch)
Now with x_train_inp any keras model can be developed.
Training (simple)
Lets say train_out is the output tensor of your keras model. You can easily write a custom training loop on the lines of:
loss = tf.reduce_mean(categorical_crossentropy(y_train_batch, train_out))
train_op = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
# sess.run(tf.global_variables_initializer())
sess.run(tf.initialize_all_variables())
with sess.as_default():
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
step = 0
while not coord.should_stop():
start_time = time.time()
_, loss_value = sess.run([train_op, loss], feed_dict={K.learning_phase(): 0})
duration = time.time() - start_time
if step % 100 == 0:
print('Step %d: loss = %.2f (%.3f sec)' % (step, loss_value,
duration))
step += 1
except tf.errors.OutOfRangeError:
print('Done training for %d epochs, %d steps.' % (FLAGS.num_epochs, step))
finally:
coord.request_stop()
coord.join(threads)
sess.close()
Training (keras style)
One of the features of keras that makes it so lucrative is its generalized training mechanism with the callback functions.
But to support tfrecords type training there are several changes that are need in the fit function
running the queue threads
no feeding in batch data through feed_dict
supporting validation becomes tricky as the validation data will also be coming in through another tensor an different model needs to be internally created with shared upper layers and validation tensor fed in by other tfrecord reader.
But all this can be easily supported by another flag parameter. What makes things messing are the keras features sample_weight and class_weight they are used to weigh each sample and weigh each class. For this in compile() keras creates placeholders (here) and placeholders are also implicitly created for the targets (here) which is not needed in our case the labels are already fed in by tfrecord readers. These placeholders needs to be fed in during session run which is unnecessary in our cae.
So taking into account these changes, compile_tfrecord(here) and fit_tfrecord(here) are the extension of compile and fit and shares say 95% of the code.
They can be used in the following way:
import keras_tfrecord as ktfr
train_model = Model(input=x_train_inp, output=train_out)
ktfr.compile_tfrecord(train_model, optimizer='rmsprop', loss='categorical_crossentropy', out_tensor_lst=[y_train_batch], metrics=['accuracy'])
train_model.summary()
ktfr.fit_tfrecord(train_model, X_train.shape[0], batch_size, nb_epoch=3)
train_model.save_weights('saved_wt.h5')
You are welcome to improve on the code and pull requests.
Update 2018-08-29 this is now directly supported in keras, see the following example:
https://github.com/keras-team/keras/blob/master/examples/mnist_tfrecord.py
Original Answer:
TFRecords are supported by using an external loss. Here are the key lines constructing an external loss:
# tf yield ops that supply dataset images and labels
x_train_batch, y_train_batch = read_and_decode_recordinput(...)
# create a basic cnn
x_train_input = Input(tensor=x_train_batch)
x_train_out = cnn_layers(x_train_input)
model = Model(inputs=x_train_input, outputs=x_train_out)
loss = keras.losses.categorical_crossentropy(y_train_batch, x_train_out)
model.add_loss(loss)
model.compile(optimizer='rmsprop', loss=None)
Here is an example for Keras 2. It works after applying the small patch #7060:
'''MNIST dataset with TensorFlow TFRecords.
Gets to 99.25% test accuracy after 12 epochs
(there is still a lot of margin for parameter tuning).
'''
import os
import copy
import time
import numpy as np
import tensorflow as tf
from tensorflow.python.ops import data_flow_ops
from keras import backend as K
from keras.models import Model
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import Flatten
from keras.layers import Input
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.callbacks import EarlyStopping
from keras.callbacks import TensorBoard
from keras.objectives import categorical_crossentropy
from keras.utils import np_utils
from keras.utils.generic_utils import Progbar
from keras import callbacks as cbks
from keras import optimizers, objectives
from keras import metrics as metrics_module
from keras.datasets import mnist
if K.backend() != 'tensorflow':
raise RuntimeError('This example can only run with the '
'TensorFlow backend for the time being, '
'because it requires TFRecords, which '
'are not supported on other platforms.')
def images_to_tfrecord(images, labels, filename):
def _int64_feature(value):
return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
def _bytes_feature(value):
return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
""" Save data into TFRecord """
if not os.path.isfile(filename):
num_examples = images.shape[0]
rows = images.shape[1]
cols = images.shape[2]
depth = images.shape[3]
print('Writing', filename)
writer = tf.python_io.TFRecordWriter(filename)
for index in range(num_examples):
image_raw = images[index].tostring()
example = tf.train.Example(features=tf.train.Features(feature={
'height': _int64_feature(rows),
'width': _int64_feature(cols),
'depth': _int64_feature(depth),
'label': _int64_feature(int(labels[index])),
'image_raw': _bytes_feature(image_raw)}))
writer.write(example.SerializeToString())
writer.close()
else:
print('tfrecord %s already exists' % filename)
def read_and_decode_recordinput(tf_glob, one_hot=True, classes=None, is_train=None,
batch_shape=[1000, 28, 28, 1], parallelism=1):
""" Return tensor to read from TFRecord """
print 'Creating graph for loading %s TFRecords...' % tf_glob
with tf.variable_scope("TFRecords"):
record_input = data_flow_ops.RecordInput(
tf_glob, batch_size=batch_shape[0], parallelism=parallelism)
records_op = record_input.get_yield_op()
records_op = tf.split(records_op, batch_shape[0], 0)
records_op = [tf.reshape(record, []) for record in records_op]
progbar = Progbar(len(records_op))
images = []
labels = []
for i, serialized_example in enumerate(records_op):
progbar.update(i)
with tf.variable_scope("parse_images", reuse=True):
features = tf.parse_single_example(
serialized_example,
features={
'label': tf.FixedLenFeature([], tf.int64),
'image_raw': tf.FixedLenFeature([], tf.string),
})
img = tf.decode_raw(features['image_raw'], tf.uint8)
img.set_shape(batch_shape[1] * batch_shape[2])
img = tf.reshape(img, [1] + batch_shape[1:])
img = tf.cast(img, tf.float32) * (1. / 255) - 0.5
label = tf.cast(features['label'], tf.int32)
if one_hot and classes:
label = tf.one_hot(label, classes)
images.append(img)
labels.append(label)
images = tf.parallel_stack(images, 0)
labels = tf.parallel_stack(labels, 0)
images = tf.cast(images, tf.float32)
images = tf.reshape(images, shape=batch_shape)
# StagingArea will store tensors
# across multiple steps to
# speed up execution
images_shape = images.get_shape()
labels_shape = labels.get_shape()
copy_stage = data_flow_ops.StagingArea(
[tf.float32, tf.float32],
shapes=[images_shape, labels_shape])
copy_stage_op = copy_stage.put(
[images, labels])
staged_images, staged_labels = copy_stage.get()
return images, labels
def save_mnist_as_tfrecord():
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train = X_train[..., np.newaxis]
X_test = X_test[..., np.newaxis]
images_to_tfrecord(images=X_train, labels=y_train, filename='train.mnist.tfrecord')
images_to_tfrecord(images=X_test, labels=y_test, filename='test.mnist.tfrecord')
def cnn_layers(x_train_input):
x = Conv2D(32, (3, 3), activation='relu', padding='valid')(x_train_input)
x = Conv2D(64, (3, 3), activation='relu')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Dropout(0.25)(x)
x = Flatten()(x)
x = Dense(128, activation='relu')(x)
x = Dropout(0.5)(x)
x_train_out = Dense(classes,
activation='softmax',
name='x_train_out')(x)
return x_train_out
sess = tf.Session()
K.set_session(sess)
save_mnist_as_tfrecord()
batch_size = 100
batch_shape = [batch_size, 28, 28, 1]
epochs = 3000
classes = 10
parallelism = 10
x_train_batch, y_train_batch = read_and_decode_recordinput(
'train.mnist.tfrecord',
one_hot=True,
classes=classes,
is_train=True,
batch_shape=batch_shape,
parallelism=parallelism)
x_test_batch, y_test_batch = read_and_decode_recordinput(
'test.mnist.tfrecord',
one_hot=True,
classes=classes,
is_train=True,
batch_shape=batch_shape,
parallelism=parallelism)
x_batch_shape = x_train_batch.get_shape().as_list()
y_batch_shape = y_train_batch.get_shape().as_list()
x_train_input = Input(tensor=x_train_batch, batch_shape=x_batch_shape)
x_train_out = cnn_layers(x_train_input)
y_train_in_out = Input(tensor=y_train_batch, batch_shape=y_batch_shape, name='y_labels')
cce = categorical_crossentropy(y_train_batch, x_train_out)
train_model = Model(inputs=[x_train_input], outputs=[x_train_out])
train_model.add_loss(cce)
train_model.compile(optimizer='rmsprop',
loss=None,
metrics=['accuracy'])
train_model.summary()
tensorboard = TensorBoard()
# tensorboard disabled due to Keras bug
train_model.fit(batch_size=batch_size,
epochs=epochs) # callbacks=[tensorboard])
train_model.save_weights('saved_wt.h5')
K.clear_session()
# Second Session, pure Keras
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train = X_train[..., np.newaxis]
X_test = X_test[..., np.newaxis]
x_test_inp = Input(batch_shape=(None,) + (X_test.shape[1:]))
test_out = cnn_layers(x_test_inp)
test_model = Model(inputs=x_test_inp, outputs=test_out)
test_model.load_weights('saved_wt.h5')
test_model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
test_model.summary()
loss, acc = test_model.evaluate(X_test, np_utils.to_categorical(y_test), classes)
print('\nTest accuracy: {0}'.format(acc))
I've also been working to improve the support for TFRecords in the following issue and pull request:
#6928 Yield Op support: High Performance Large Datasets via TFRecords, and RecordInput
#7102 Keras Input Tensor API Design Proposal
Finally, it is possible to use tf.contrib.learn.Experiment to train Keras models in TensorFlow.