def load_image(path):
'''Method to load the image'''
img = image.load_img(path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
return np.asarray(x)
def Standard(path, size):
img = image.load_img(path, target_size = (size, size))
x = image.img_to_array(img)
x /= 255.0
x -= 0.5
x *= 2.0
return x
def Standard(path, size):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
x /= 255.0
x -= 0.5
x *= 2.0
return x
def Normalize(path, size=224, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
x /= 255.0
for i in range(0, 3):
x[..., i] -= mean[i]
x[..., i] /= std[i]
return x
def read_and_prepare_images(img_paths, img_height=img_size, img_width=img_size):
# image loading via load_image
imgs = [load_img(img_path, target_size=(img_height, img_width)) for img_path in img_paths]
# for storing image in 3-D tensors: img_to_array
img_array = np.array([img_to_array(img) for img in imgs])
# preprocess_input applies arithmetic on pixels so that their value is b/w -1 and 1
output = preprocess_input(img_array)
return output
def Normalize(path, size=224, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
x /= 255.0
for i in range(0, 3):
x[..., i] -= mean[i]
x[..., i] /= std[i]
return x
def ZeroCenter(path, size, BGRTranspose=False):
img = image.load_img(path, target_size = (size, size))
x = image.img_to_array(img)
if BGRTranspose == True:
x = x[..., ::-1]
x[..., 0] -= 103.939
x[..., 1] -= 116.779
x[..., 2] -= 123.68
return x
def ZeroCenter(path, size, BGRTranspose=False):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
if BGRTranspose == True:
x = x[..., ::-1]
x[..., 0] -= 103.939
x[..., 1] -= 116.779
x[..., 2] -= 123.68
return x
def Standard(path, size, BGRTranspose=False):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
x /= 255.0
x -= 0.5
x *= 2.0
if BGRTranspose == True:
x = x[..., ::-1]
return x
def Normalize(path, size=224, scale=0.0392156863 ,mean=[-0.485, -0.456, -0.406], std=[0.229, 0.224, 0.225], BGRTranspose = False):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
x *= scale
for i in range(0, 3):
x[..., i] += mean[i]
x[..., i] /= std[i]
if BGRTranspose == True:
x = x[..., ::-1]
return x
def preprocess(img_path):
# Preprocessing
img = load_img(img_path)
img = img_to_array(img)
#Tensorflow backend dimensions
img = imresize(img, (img_h, img_w, 3))
img = img.astype('float64')
#Batch dimension
img = np.expand_dims(img, axis=0)
img = vgg19.preprocess_input(img)
return img
def preprocess(img_path):
# Preprocessing to make the style transfer
# possible
img = load_img(img_path)
img = img_to_array(img)
# This dimensions are for Tensorflow backend
img = imresize(img, (img_h, img_w, 3))
img = img.astype('float64')
# Add the batch dimension
img = np.expand_dims(img, axis=0)
img = vgg19.preprocess_input(img)
return img
def ZeroCenter(path, size, BGRTranspose=False):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
# Reference: 1) Keras image preprocess: https://github.com/keras-team/keras/blob/master/keras/applications/imagenet_utils.py
# 2) tensorflow github issue: https://github.com/tensorflow/models/issues/517
# R-G-B for Imagenet === [123.68, 116.78, 103.94]
x[..., 0] -= 123.68
x[..., 1] -= 116.779
x[..., 2] -= 103.939
if BGRTranspose == True:
x = x[..., ::-1]
return x
def __data_label__(path):
datalist = []
labellist = []
for i in range(0, 7500):
imgname = trainpath + str('%04d' % i) + ".jpg"
f = open(trainpath + str('%04d' % i) + ".pts", "r")
new_label = []
for line in f.readlines():
a = line.replace("\n", "")
b = a.split(",")
new_label.append(b[0])
new_label.append(b[1])
labellist.append(new_label)
image = load_img(trainpath + str('%04d' % i) + ".jpg")
datalist.append(np.array(img_to_array(image).tolist()))
img_data = np.array(datalist)
img_data /= 255
label = np.array(labellist)
print(img_data)
return img_data, label
from tensorflow.contrib.keras.api.keras.preprocessing import image
import tensorflow as tf
assert tf.__version__ == "1.8.0"
tf.set_random_seed(20180130)
img_path = 'sample.jpg'
import numpy as np
x = image.load_img(img_path, target_size=(250, 250))
x = image.img_to_array(x)
x_expended = np.expand_dims(x, axis=0)
x_expended_trans = np.transpose(x_expended, [0, 3, 1, 2])
X = tf.placeholder(tf.float32, [None, 250, 250, 3])
sess = tf.Session()
sess.run(tf.global_variables_initializer())
print(sess.run(X, feed_dict={X: x_expended_trans}))
validation_steps=2000 / batch_size,
workers=12,
max_q_size=100,
callbacks=[history])
# Serialize Model
ModelSerializer.serialize_model_json(classifier, 'loss_history',
'loss_history_weights')
# Predict single cases
test_image_1 = image.load_img('dataset/single_prediction/cat_or_dog_1.jpg',
target_size=input_size)
test_image_2 = image.load_img('dataset/single_prediction/cat_or_dog_2.jpg',
target_size=input_size)
test_image_1 = image.img_to_array(test_image_1)
test_image_2 = image.img_to_array(test_image_2)
# adding a 4th dimension for predict method
# this dimension corresponds to the batch, cannot accept single inputs
# only batch of size single or greater inputs
test_image_1 = np.expand_dims(test_image_1, axis=0)
test_image_2 = np.expand_dims(test_image_2, axis=0)
predict_1 = classifier.predict(test_image_1)
predict_2 = classifier.predict(test_image_2)
# to understand output of 0 or 1, cats are 0, dogs are 1
training_set.class_indices
if predict_1[0][0] == 1:
import os
script_dir = os.path.dirname(__file__)
# Load pre-trained model
model_backup_path = os.path.join(script_dir, '../dataset/cat_or_dogs_model.h5')
test_set_path = os.path.join(script_dir, '../dataset/single_prediction')
classifier = load_model(model_backup_path)
input_size = (128, 128)
test_images_path = [
test_set_path + '/' + filename for filename in os.listdir(test_set_path)
]
test_images = np.array([
image.img_to_array(image.load_img(test_image_name, target_size=input_size))
for test_image_name in test_images_path
])
# No need to rescale the images here... why?
predictions = classifier.predict(test_images)
for prediction, image_path in zip(predictions, test_images_path):
if prediction == 1:
prediction = 'dog'
else:
prediction = 'cat'
print("Predicted {} for file {}".format(prediction,
image_path.split("/")[-1]))
self.losses = losses
return loss
if __name__ == "__main__":
import sys
from tensorflow.contrib.keras.api.keras.preprocessing import (
image)
s = tf.Session()
img_path = sys.argv[1]
img = image.load_img(img_path, target_size=(128, 128))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = x / 255.0 * 2.0 - 1.0
print(x.shape, np.min(x), np.max(x))
x = tf.constant(x)
feature_layers = [
"input_1", "block1_conv1", "block1_conv2", "block1_pool", "block2_conv2",
"block3_conv2", "block4_conv2", "block5_conv2"]
vgg19 = VGG19Features(feature_layers)
fmaps = vgg19.make_feature_ops(x)
for i in range(len(fmaps)):
print(i)
f = fmaps[i].eval(session=s)
print(f.shape)
def Identity(path, size, BGRTranspose=False):
img = image.load_img(path, target_size = (size, size))
x = image.img_to_array(img)
if BGRTranspose == True:
x = x[..., ::-1]
return x
def Identity(path, size, BGRTranspose=False):
img = image.load_img(path, target_size=(size, size))
x = image.img_to_array(img)
if BGRTranspose == True:
x = x[..., ::-1]
return x
def generate_example(path, group):
img_path = ['{}_group{}_index{}.jpg'.format(path, group, idx) for idx in range(time_step)]
imgs = [image.img_to_array(image.load_img(x)) for x in img_path]
return imgs