def evaluate(self, dataset_path):
pil_images, image_classes, num_samples, num_classes = load_image_classification_dataset(dataset_path, 'RGB')
images = self._trans_image(pil_images)
# images = np.pad(images, ((0, 0), (2, 2), (2, 2), (0, 0)), 'constant')
X_test, y_test = images, to_categorical(image_classes, num_classes=self._num_classes)
# Compute test accuracy
(test_loss, test_acc) = self._model.evaluate(X_test, y_test)
return test_loss, test_acc
def train(self, dataset_path, **train_args):
batch_size = train_args.get('batch_size')
max_epochs = train_args.get('max_epochs')
learning_rate = float(train_args['learning_rate'])
pil_images, image_classes, num_samples, num_classes = load_image_classification_dataset(
dataset_path, 'RGB')
self._num_classes = num_classes
images = self._trans_image(pil_images)
# pil_images = [x.resize([self._image_size, self._image_size]) for x in pil_images]
# # Convert to numpy arrays
# images = [np.asarray(x) for x in pil_images]
# images = np.asarray(images) / 255
# images = self._prepare_X(images)
train = {}
train['images'] = images
train['classes'] = image_classes
validation = {}
train['images'], validation['images'], train['classes'], validation[
'classes'] = train_test_split(train['images'],
train['classes'],
test_size=0.2,
random_state=0)
# train['images'] = np.pad(train['images'],
# ((0, 0), (2, 2), (2, 2), (0, 0)), 'constant')
# validation['images'] = np.pad(validation['images'],
# ((0, 0), (2, 2), (2, 2), (0, 0)),
# 'constant')
X_train, y_train = train['images'], to_categorical(
train['classes'], num_classes=num_classes)
X_validation, y_validation = validation['images'], to_categorical(
validation['classes'], num_classes=self._num_classes)
train_generator = ImageDataGenerator().flow(X_train,
y_train,
batch_size=batch_size)
validation_generator = ImageDataGenerator().flow(X_validation,
y_validation,
batch_size=batch_size)
steps_per_epoch = X_train.shape[0] // batch_size
validation_steps = X_validation.shape[0] // batch_size
tensorboard = TensorBoard(log_dir="logs/{}".format(time()))
if not self._model:
self._model = self._build_classifier(float(learning_rate))
logger.log(logging.INFO, 'Begin loading train dataset')
self._model.fit_generator(train_generator,
steps_per_epoch=steps_per_epoch,
epochs=max_epochs,
validation_data=validation_generator,
validation_steps=validation_steps,
shuffle=True,
callbacks=[tensorboard])
# Compute train accuracy
(train_loss, train_acc) = self._model.evaluate(X_validation,
y_validation)
logger.info('Train loss: {}'.format(train_loss))
logger.info('Train accuracy: {}'.format(train_acc))