if kernel_type == "gaussian":
clf = my_SVM.my_SVM_train("gaussian",
my_SVM.my_Kernel.gaussian(50))
else:
clf = my_SVM.my_SVM_train("linear", my_SVM.my_Kernel.linear())
_clf = clf.train(np.asarray(trainingX), np.asarray(trainingY))
average += _clf.score(_clf.predict(np.asarray(testX)),
np.asarray(testY))
return average / 100
# Loading Images
dataset = load_images.load()
# Extracting Features
corner_dataset = corner_feature.corner_feature(dataset)
edge_dataset = edge_feature.edge_feature(load_images.load())
hog_dataset = hog_feature.hog_feature(load_images.load())
# Corner
# Linear Kernel -> Accuracy 64.26%
print(my_classifier("linear", corner_dataset))
# Gaussian Kernel -> Accuracy 52.41%
print(my_classifier("gaussian", corner_dataset))
from model_architecture import Model
from sklearn.model_selection import train_test_split
from keras.utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
epochs = 25
learning_rate = 0.001
bs = 32
data = []
labels = []
architecture = Model(64, 64, 3, 3)
model = architecture.model3()
model.summary()
ld = load(64, 64, 3,
[['D:/blind project 2/left'], ['D:/blind project 2/right'],
['D:/blind project 2/center']])
data, labels = ld.imgload()
# print(data)
c = list(zip(data, labels))
random.shuffle(c)
data[:], labels[:] = zip(*c)
data = np.array(data, dtype="float") / 255.0
labels = np.array(labels)
trainX, testX, trainY, testY = train_test_split(data,
labels,
test_size=0.1,
random_state=42)
trainY = to_categorical(trainY, num_classes=3)
testY = to_categorical(testY, num_classes=3)
from load_images import load from matplotlib import pyplot import numpy as np X, y = load() avg_kaggle_image = np.zeros((96, 96)) for i in range(X.shape[0]): avg_kaggle_image += X[i].reshape(96, 96) avg_kaggle_image /= X.shape[0] fig = pyplot.figure(figsize=(1, 1)) ax = fig.add_subplot(1, 1, 1, xticks=[], yticks=[]) ax.imshow(avg_kaggle_image, cmap="gray") pyplot.show()
clip_disc_weights = None
# For saving samples
fixed_noise = tf.constant(np.random.normal(size=(128, 128)).astype('float32'))
fixed_noise_samples = Generator(128, noise=fixed_noise)
def generate_image(frame, true_dist):
samples = session.run(fixed_noise_samples)
lib.save_images.save_images(samples.reshape((128, 28, 28)),
out_dir + 'samples_{}.png'.format(frame))
# Dataset iterator
#train_gen, dev_gen, test_gen = lib.mnist.load(BATCH_SIZE, BATCH_SIZE)
train_gen, dev_gen = loader.load(BATCH_SIZE, BATCH_SIZE, imarr_fn)
def inf_train_gen():
while True:
for images in train_gen():
yield images
print("Training")
# Train loop
with tf.Session() as session:
session.run(tf.initialize_all_variables())
gen = inf_train_gen()
x = np.append(px, nx, axis=0)
y = np.append(py, ny)
return (x, y)
def train_svm(feature, label):
clf = AdaBoostClassifier(svm.SVC(probability=True,kernel='linear'),n_estimators=20)
clf.fit(feature, label)
return clf
if __name__ == "__main__":
minWinSize = [64, 64]
p_images = li.load("positive_train_images")
n_images = li.load("negative_train_images")
p_test_images = li.load("positive_test_images")
n_test_images = li.load("negative_test_images")
test_images = np.append(p_test_images, n_test_images, axis=0)
px = np.array(extract_hog(p_images))
nx = np.array(extract_hog(n_images))
py = np.ones(len(p_images))
ny = np.zeros(len(n_images))
test_py = np.ones(len(p_test_images))
test_ny = np.zeros(len(n_test_images))
def model_main():
start = time.time()
tf.logging.set_verbosity(tf.logging.DEBUG)
_logger = logging.getLogger("tensorflow")
_logger.info("--------------------- Load Kubernetes Config ---------------------")
tf_config = kubernetes_resolver.build_config()
os.environ['TF_CONFIG'] = str(tf_config)
worker_index = kubernetes_resolver.fetch_task_index()
num_workers = len(kubernetes_resolver.build_worker_list())
# Local setup
#
# worker_index = None
# num_workers = 3
_logger.info("--------------------- Load Data ---------------------")
(x_train, y_train), (x_test, y_test) = load_images.load()
_logger.info("--------------------- Set RunConfiguration ---------------------")
distribution = tf.contrib.distribute.CollectiveAllReduceStrategy(num_gpus_per_worker=0)
config = tf.estimator.RunConfig(train_distribute=distribution,
eval_distribute=distribution)
# Local setup
#
# config = None
# Create estimator
_logger.info("--------------------- Create Estimator ---------------------")
keras_estimator = tf.keras.estimator.model_to_estimator(
keras_model=create_model(), config=config, model_dir='./model')
train_spec = tf.estimator.TrainSpec(
input_fn=lambda: input_fn(img=x_train, label=y_train,
num_workers=num_workers,
worker_index=worker_index,
shuffle=True), max_steps=math.floor(1000 / num_workers))
eval_spec = tf.estimator.EvalSpec(
input_fn=lambda: input_fn(img=x_test, label=y_test,
num_workers=num_workers,
worker_index=worker_index,
shuffle=False), steps=100)
# Create estimator
tf.LogMessage()
_logger.info("--------------------- Start Training ---------------------")
tf.estimator.train_and_evaluate(keras_estimator, train_spec, eval_spec)
_logger.info("--------------------- Finish training ---------------------")
end = time.time()
time_diff = end - start
_logger.info('--------------------- Estimate time ---------------------')
_logger.info('Tensorflow Time start: {}'.format(start))
_logger.info('Tensorflow Time end: {}'.format(end))
_logger.info('Tensorflow Time elapased: {}'.format(time_diff))
_logger.info("--------------------- Start Export ---------------------")
export_dir = keras_estimator.export_savedmodel(
export_dir_base="./dist",
serving_input_receiver_fn=serving_input_fn)
_logger.info("--------------------- Finish Export on Path %s ---------------------"
% export_dir)
_logger.info("--------------------- Start Tensorboard ---------------------")
if "TF_CONFIG" in os.environ:
config = os.environ['TF_CONFIG']
if "\"type\": \"chief\"" in config:
os.system('tensorboard --logdir=/notebooks/app/model --port=6006')
from model_architecture import Model
from sklearn.model_selection import train_test_split
from keras.utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
epochs = 25
learning_rate = 0.001
bs = 32
data = []
labels = []
architecture = Model(64, 64, 3, 3)
model = architecture.model3()
model.summary()
ld = load(64, 64, 3,
[['C:\Users\ROUSHAN K\Desktop\MDD_PROJECT\SMART_SPECS\LEFT'],
['C:\Users\ROUSHAN K\Desktop\MDD_PROJECT\SMART_SPECS\RIGHT'],
['C:\Users\ROUSHAN K\Desktop\MDD_PROJECT\SMART_SPECS\CENTRE']])
data, labels = ld.imgload()
# print(data)
c = list(zip(data, labels))
random.shuffle(c)
data[:], labels[:] = zip(*c)
data = np.array(data, dtype="float") / 255.0
labels = np.array(labels)
trainX, testX, trainY, testY = train_test_split(data,
labels,
test_size=0.1,
random_state=42)
trainY = to_categorical(trainY, num_classes=3)
testY = to_categorical(testY, num_classes=3)
parser.add_argument('--gpu', '-g', default=-1, type=int,
help='GPU ID (negative value indicates CPU)')
args = parser.parse_args()
if args.gpu >= 0:
cuda.check_cuda_available()
xp = cuda.cupy if args.gpu >= 0 else np
batchsize = 100
n_epoch = 5
resize = 50
predict_no = 80
#load train data
print('load Sunshine dataset')
x_train, x_test, y_train, y_test = load_images.load("./image",11,10,resize,resize)
cv2.imshow("predict",x_test[predict_no])
cv2.waitKey(0)
#Normalization 0~1
x_train = np.array(x_train).astype(np.float32).reshape((len(x_train),3, resize, resize)) / 255
y_train = np.array(y_train).astype(np.int32)
x_test = np.array(x_test).astype(np.float32).reshape((len(x_test),3, resize, resize)) / 255
y_test = np.array(y_test).astype(np.int32)
N = len(y_train)
print(N)
print(len(x_test))
N_test = y_test.size
import imageio
import matplotlib.pyplot as plt
import numpy as np
import os
import PIL
from tensorflow.keras import layers
import time
import load_images
from IPython import display
# this is the code that trains and saves GANs on my images
# the code itself was mostly copied from https://www.tensorflow.org/tutorials/generative/dcgan
# I had to modify the architectures of the generator and discriminator and change the training snapshot code
# because my images were RGB and had larger dimensions than the data the example was written for
train_images = load_images.load(num=-1)
BUFFER_SIZE = len(train_images)
BATCH_SIZE = 64
# Batch and shuffle the data
train_dataset = tf.data.Dataset.from_tensor_slices(train_images).shuffle(
BUFFER_SIZE).batch(BATCH_SIZE)
# This is where the generator's layer structure is defined
# I had to change this significantly
def make_generator_model():
model = tf.keras.Sequential()
model.add(layers.Dense(8 * 5 * 256, use_bias=False, input_shape=(100, )))
model.add(layers.BatchNormalization())
import numpy as np
from model_architecture import Model
from sklearn.model_selection import train_test_split
from keras.utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
epochs = 25
learning_rate = 0.001
bs = 32
data = []
labels = []
architecture = Model(64, 64, 3, 3)
model = architecture.model3()
model.summary()
ld = load(64, 64, 3, [['D:/blind project 2/left'], ['D:/blind project 2/right'], ['D:/blind project 2/center']])
data, labels = ld.imgload()
# print(data)
c = list(zip(data, labels))
random.shuffle(c)
data[:], labels[:] = zip(*c)
data = np.array(data, dtype="float")/255.0
labels = np.array(labels)
trainX, testX, trainY, testY = train_test_split(data, labels, test_size=0.1, random_state=42)
trainY = to_categorical(trainY, num_classes=3)
testY = to_categorical(testY, num_classes=3)
aug = ImageDataGenerator(rotation_range=30, width_shift_range=0.1, height_shift_range=0.1, shear_range=0.2,
zoom_range=0.2, fill_mode="nearest")
opt = Adam(lr=learning_rate, decay=learning_rate/epochs)
model.compile(loss="categorical_crossentropy", optimizer=opt, metrics=["accuracy"])