def test_predictions():
# prepare test data
image_data_generator = ImageDataGenerator(rescale=(1.0 / 255))
# load test dog images
test_dog_image_list = glob.glob(DEFAULT_TEST_DATA_DIR + "/dogs/dog*.jpg")
test_dog_images = []
for dog_file in test_dog_image_list:
dog_image = image.load_img(dog_file,
target_size=(img_height, img_width))
rescaled_dog_image = image_data_generator.standardize(
image.img_to_array(dog_image))
test_dog_images.append(rescaled_dog_image)
# load test cat images
test_cat_image_list = glob.glob(DEFAULT_TEST_DATA_DIR + "/cats/cat*.jpg")
test_cat_images = []
for cat_file in test_cat_image_list:
cat_image = image.load_img(cat_file,
target_size=(img_height, img_width))
rescaled_cat_image = image_data_generator.standardize(
image.img_to_array(cat_image))
test_cat_images.append(rescaled_cat_image)
# load trained cnn model
cnn_model = load_cnn_model(DEFAULT_MODEL_FILE)
print("\nTest Trained CNN Model with Dog Images: =======================")
for dog_image in test_dog_images:
prediction = cnn_model.predict(np.array([dog_image]), batch_size=1)
print(" Predicted Probability of Dog Image is {}".format(
prediction[0]))
print("\nTest Trained CNN Model with Cat Images: =======================")
for cat_image in test_cat_images:
prediction = cnn_model.predict(np.array([cat_image]), batch_size=1)
print(" Predicted Probability of Dog Image is {}".format(
prediction[0]))
import numpy as np
import os
from time import process_time
from time import process_time_ns
#test batch predikce
start = process_time()
img_dir = "K:/data/manual test"
batch = np.zeros((len(os.listdir(img_dir)), 256, 256, 3))
for i, img in enumerate(os.listdir(img_dir)):
loadedimg = image.load_img(os.path.join(img_dir, img),
target_size=(256, 256))
batch[i, ] = loadedimg
batch = datagen.standardize(batch)
predictions = model.predict_proba(batch, batch_size=len(
os.listdir(img_dir))) #.tolist()[0]
finalPreds = []
for prediction in predictions:
print(prediction)
dictionary = {}
for label in class_names:
class_name = class_names.get(label)
dictionary.update({class_name: prediction[label]})
finalPreds.append(dictionary)
assignDict = {}
filelist = os.listdir(img_dir)
for i, pred in enumerate(finalPreds):
print(i)
assignDict.update({filelist[i]: pred})
y_train, y_test_low, y_test_high, y_test_random, X_train, X_test_low, X_test_high, X_test_random = \
train_test_many_split(y, X, groups=data['ID'], side_test_size=0.05, random_test_size=0.1)
# Get and train data generator
# TODO: solid angle rotations
datagen = ImageDataGenerator(
featurewise_center=True,
featurewise_std_normalization=True,
# rotation_range=30,
horizontal_flip=True,
vertical_flip=True,
)
datagen.fit(X_train)
# Standardize validation datasets
X_test_random = datagen.standardize(X_test_random)
X_test_high = datagen.standardize(X_test_high)
X_test_low = datagen.standardize(X_test_low)
# Train model
# TODO: send some images to tensorboard
callbacks = [
CustomTensorBoard(log_dir=get_tensorboard_dir(args, timestamp,
learning_rate, batch_size),
validation_data={
'low': (X_test_low, y_test_low),
'high': (X_test_high, y_test_high)
}),
EarlyStopping(monitor='val_loss',
patience=patience,
restore_best_weights=True),
