def load_mnist(version, preprocess, categorical_labels=True, custom_name=''):
num_classes = 10
image_size = 28
model_params = dict()
model_params['num_classes'] = num_classes
custom_name = '' if str(custom_name) == '' else str(custom_name) + '_'
model_params['name'] = 'MLP_MNIST_' + str(custom_name) + 'v' + str(version)
(x_train, y_train), (x_test, y_test) = mnist.load_data()
if preprocess == 'line_profile':
model_params['input_dim'] = image_size * 2
lg_28 = laguerre_gauss_filter(image_size, 0.9)
ft_lg_28 = np.fft.fft2(lg_28)
x_pr_train, y_pr_train = ft_pipeline(ft_lg_28, x_train)
x_pr_test, y_pr_test = ft_pipeline(ft_lg_28, x_test)
x_train = np.abs(np.concatenate((x_pr_train, y_pr_train), axis=1))
x_test = np.abs(np.concatenate((x_pr_test, y_pr_test), axis=1))
elif preprocess == 'flattened':
x_train = x_train.reshape(x_train.shape[0], -1)
x_test = x_test.reshape(x_test.shape[0], -1)
model_params['input_dim'] = image_size**2
else:
img_rows, img_cols = image_size, image_size
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
model_params['input_shape'] = input_shape
model_params['name'] = 'CNN_MNIST_' + str(custom_name) + 'v' + str(
version)
if categorical_labels:
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
splits = [('train', x_train, y_train), ('validation', x_test, y_test)]
data = dict()
for spl in splits:
split, features, labels = spl
model_params[split + '_imbalance'] = np.unique(labels,
axis=0,
return_counts=True)
data[split] = (features, labels)
return model_params, data
def get_line_profiles(x, size, omega=0.9):
lg_filter = laguerre_gauss_filter(size, omega)
ft_lg_filter = np.fft.fft2(lg_filter)
x_profile, y_profile = ft_pipeline(ft_lg_filter, x)
# ft_pipeline_no_shift(ft_lg_filter, x)
# ft_pipeline_no_transform(ft_lg_filter, x)
return np.abs(np.concatenate((x_profile, y_profile), axis=1))
def run(self, data_path, video=False):
data = find_files(data_path, sorted_by_idx=False)
if video:
media = load_videos(data, data_path)
else:
media = load_images(data, data_path)
if video:
print(len(media), len(media[0]))
size = self.model_specs['size']
omega = self.model_specs['omega']
cropper = CropImage(size)
for instance in media:
instance = rgb2grayscale(instance)
plt.imshow(instance, cmap='gray')
plt.title('Input')
plt.show()
original_crops = cropper.crop(instance)
lg_filter = laguerre_gauss_filter(size, omega)
ft_lg_filter = np.fft.fft2(lg_filter)
x_profile, y_profile = ft_pipeline(ft_lg_filter, original_crops)
ft_crops = np.abs(np.concatenate((x_profile, y_profile), axis=1))
if not self.is_joblib:
predictions = self.model.predict_on_batch(ft_crops)
else:
predictions = self.model.predict(ft_crops)
print(predictions)
pos_preds = []
for i in range(len(predictions)):
if not self.is_joblib:
pred = np.argmax(predictions[i])
else:
pred = predictions[i]
if pred == 0:
continue
# img = original_crops[i]
# plt.imshow(img, cmap='gray')
# plt.title('Prediction: ' + str(pred))
# plt.show()
# plt.pause()
# plt.close()
pos_preds.append(original_crops[i])
print(len(pos_preds))
n_row, n_col = 2, 3
_, axs = plt.subplots(n_row, n_col, figsize=(12, 12))
axs = axs.flatten()
plt.suptitle('Positive Predictions kNN')
for img, ax in zip(pos_preds, axs):
ax.imshow(img, cmap='gray')
plt.show()
from tensorflow.keras.layers import Dense, Dropout import numpy as np # ------------ LOAD DATA ----------------------- (x_train, y_train), (x_test, y_test) = mnist.load_data() # ------------ PRE-PROCESS DATA ----------------------- image_size = 28 lg_28 = laguerre_gauss_filter(image_size, 0.9) ft_lg_28 = np.fft.fft2(lg_28) x_pr_train, y_pr_train = ft_pipeline(ft_lg_28, x_train) x_pr_test, y_pr_test = ft_pipeline(ft_lg_28, x_test) x_train = np.concatenate((x_pr_train, y_pr_train), axis=1) x_test = np.concatenate((x_pr_test, y_pr_test), axis=1) num_classes = 10 # convert class vectors to binary class matrices y_train = keras.utils.to_categorical(y_train, num_classes) y_test = keras.utils.to_categorical(y_test, num_classes) # ------------ MODEL ----------------------- model = Sequential() model.add(Dense(256, activation='relu', input_dim=56, name='my1'))
data_augmentation = True
num_predictions = 20
save_dir = os.path.join(os.getcwd(), 'saved_models')
model_name = 'keras_cifar10_trained_model.h5'
# ------------ LOAD DATA -----------------------
# The data, split between train and test sets:
(x_train_org, y_train_org), (x_test_org, y_test_org) = cifar10.load_data()
# ------------ PRE-PROCESS DATA -----------------------
lg_filter_32 = laguerre_gauss_filter(32, 0.9)
ft_lg_32 = np.fft.fft2(lg_filter_32)
x_pr_train, y_pr_train = ft_pipeline(ft_lg_32, x_train_org)
x_pr_test, y_pr_test = ft_pipeline(ft_lg_32, x_test_org)
x_train = np.abs(np.concatenate((x_pr_train, y_pr_train), axis=1))
x_test = np.abs(np.concatenate((x_pr_test, y_pr_test), axis=1))
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# Convert class vectors to binary class matrices.
y_train = keras.utils.to_categorical(y_train_org, num_classes)
y_test = keras.utils.to_categorical(y_test_org, num_classes)
# ------------ MODEL -----------------------
def get_line_profiles(x):
x_profile, y_profile = ft_pipeline(ft_lg_32, x)
return np.abs(np.concatenate((x_profile, y_profile), axis=1))