Created
March 20, 2018 03:12
-
-
Save yudai09/70264f3910bfc43835b920e7e9f96007 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import glob | |
import numpy as np | |
import keras | |
from keras.callbacks import ModelCheckpoint | |
from keras.preprocessing import image | |
from keras.preprocessing.image import ImageDataGenerator | |
from keras.layers import Input | |
from vis.visualization import visualize_cam, overlay | |
from matplotlib import pyplot as plt | |
def main(img_width=224, img_height=224, data_dir = './data/Stanford40/JPEGImages/'): | |
model_name = 'InceptionV4' | |
nb_classes = 40 | |
model_path = None | |
model = None | |
if model_name == 'VGG19': | |
model_path='./weight/vgg19_full.hdf5' | |
input_tensor = Input(shape=(img_height, img_width, 3)) | |
# Base model is VGG19 | |
# Pretrained by using imagenet data set | |
base_model = keras.applications.vgg19.VGG19(include_top=False, weights='imagenet', input_tensor=input_tensor) | |
# Batch normalization | |
model = Sequential() | |
model.add(BatchNormalization(input_shape=(img_width, img_height, 3))) | |
model.add(base_model) | |
# Add FC layer for designated number of classes | |
fc_model = Sequential() | |
fc_model.add(Flatten(input_shape=base_model.output_shape[1:])) | |
fc_model.add(Dense(256, activation='relu')) | |
fc_model.add(Dropout(0.5)) | |
fc_model.add(Dense(nb_classes, activation='softmax')) | |
model.add(fc_model) | |
else: # model == InceptionV4 | |
input_tensor = Input(shape=(img_height, img_width, 3)) | |
model_path='./weight/inceptionV4_full.hdf5' | |
model = keras.applications.inception_resnet_v2.InceptionResNetV2(include_top=True, weights=None, input_tensor=input_tensor, classes=nb_classes) | |
model.compile(loss='categorical_crossentropy', | |
optimizer=keras.optimizers.SGD(lr=1e-3, momentum=0.90, decay=1e-8, nesterov=True), | |
metrics=['accuracy']) | |
model.summary() | |
print(model.layers) | |
print('{} layers'.format(len(model.layers))) | |
if os.path.exists(model_path): | |
model.load_weights(model_path) | |
classes = ['applauding', 'blowing_bubbles', 'brushing_teeth', 'cleaning_the_floor', 'climbing', 'cooking', 'cutting_trees', 'cutting_vegetables', 'drinking', 'feeding_a_horse', 'fishing', 'fixing_a_bike', 'fixing_a_car', 'gardening', 'holding_an_umbrella', 'jumping', 'looking_through_a_microscope', 'looking_through_a_telescope', 'phoning', 'playing_guitar', 'playing_violin', 'pouring_liquid', 'pushing_a_cart', 'reading', 'riding_a_bike', 'riding_a_horse', 'rowing_a_boat', 'running', 'shooting_an_arrow', 'smoking', 'taking_photos', 'texting_message', 'throwing_frisby', 'using_a_computer', 'walking_the_dog', 'washing_dishes', 'watching_TV', 'waving_hands', 'writing_on_a_board', 'writing_on_a_book'] | |
data_dir='./data/Stanford40/JPEGImages/' | |
for image_path in glob.glob("{}/*/*".format(data_dir)): | |
original_image, preprocessed_input = load_image(image_path) | |
predictions = model.predict(preprocessed_input) | |
print(predictions) | |
prediction_iter = np.argmax(predictions) | |
print('{} {} {}'.format(image_path, prediction_iter, classes[prediction_iter])) | |
gradcam = visualize_cam(model, len(model.layers)-1, prediction_iter, preprocessed_input) | |
# print(gradcam) | |
print(preprocessed_input.shape) | |
print(original_image.shape) | |
print(gradcam.shape) | |
plt.imshow(overlay(gradcam, original_image)) | |
plt.show() | |
while True: | |
pass | |
def load_image(path): | |
img_path = path | |
# -----begin image.load_img----- | |
img = image.load_img(img_path, target_size=(224, 224)) | |
img = image.img_to_array(img) | |
x = np.expand_dims(img, axis=0) | |
return img, x / 255 | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment