def load_mydata(CONSTRUCTION_METHOD, NUMBER_OF_CLASSES, csv_filepath,
data_filepath):
if CONSTRUCTION_METHOD == 'image':
train_df, input_shape = data_from_csv(csv_filepath, data_filepath)
print('Create Tensor Data...')
x_train_channelb = read_images(train_df.channelb.values, input_shape)
x_train_channelc = read_images(train_df.channelc.values, input_shape)
x_train_channeld = read_images(train_df.channeld.values, input_shape)
else:
train_df, input_shape = data_from_csv_nonimage(csv_filepath,
data_filepath)
x_train_channelb = read_data(train_df.channelb.values, input_shape)
x_train_channelc = read_data(train_df.channelc.values, input_shape)
x_train_channeld = read_data(train_df.channeld.values, input_shape)
# labels - convert class vectors to binary class matrices One Hot Encoding
labels = train_df.object.values
labels = to_categorical(labels, NUMBER_OF_CLASSES)
x_train_comp = np.stack(
(x_train_channelb, x_train_channelc, x_train_channeld), axis=4)
x_train, x_test, y_train, y_test = train_test_split(x_train_comp,
labels,
test_size=0.3,
random_state=None)
return (x_train, y_train), (x_test, y_test)
def test(model, attack_func, method_name):
if os.path.exists("results"):
shutil.rmtree("results")
os.mkdir("results");
acc = 0
for (file_name, image, label) in read_images():
if test_model_acc == True:
acc += np.argmax(model.predictions(image)) == label
continue
print(file_name, end="\t\t")
np.random.seed(label+2)
target_class = int(np.random.random()*200)
adversarial = attack_func(model, image, target_class)
store_adversarial(file_name, adversarial)
if adversarial is None:
print("can't find")
elif np.argmax(model.predictions(adversarial)) == target_class:
print("l2: %.4f" %np.linalg.norm(image/255 - adversarial/255))
else:
print("error");
exit()
if test_model_acc == True:
print("model accuracy: %.4f" %(acc/200)); exit()
print("\n", method_name, "\n")
compute_MAD()
def generate(files, steering, batch_size, augment_data=True):
#data = shuffle(data)
num_examples = len(files)
offset = num_examples
i = 1
while True:
if (offset + batch_size) >= num_examples:
offset = 0
i = 1
files_s, steering_s = shuffle(files, steering)
for offset in range(0, num_examples, batch_size):
i += 1
end = offset + batch_size
if end >= num_examples:
end = num_examples
filename_x, batch_y = files_s[offset:end], steering_s[offset:end]
if augment_data:
batch_x, batch_y = utils.augment_dataset_single(
filename_x, batch_y)
else:
batch_x = utils.read_images(filename_x)
# Rescale and resize only
batch_x = utils.preprocess_images(batch_x, False)
yield batch_x.astype('float32'), batch_y.astype('float32')
def create_forest():
""" Create new random decision forest
"""
print('Creating random forest')
images = read_images()
# assign classes
image_classes = [icls for icls, _ in images]
# extract features
feature_vectors = []
for _, img_path in images:
cur = get_feature_vector(img_path)
feature_vectors.append(cur)
assert len(feature_vectors) > 0, 'Must have at least one feature vector to train on'
# good estimator number for classification tasks
sqrt_feat_num = int(np.sqrt(len(feature_vectors[0])))
print(' ', sqrt_feat_num, 'estimator%s' % ('' if sqrt_feat_num == 1 else 's'))
# create forest
clf = RandomForestClassifier(
n_estimators=sqrt_feat_num, n_jobs=-1,
max_depth=None, min_samples_split=1
)
clf = clf.fit(feature_vectors, image_classes)
return clf
def test(thetas):
size_img, rows, cols, images = utils.read_images(
TRAINING_IMAGES_PATH
)
size_lbl, labels = utils.read_labels(TRAINING_LABELS_PATH)
bias_terms = np.ones([size_img, 1])
images = np.concatenate((bias_terms, images), axis=1)
training_images = images[range(0, M_TRAINING), :]
training_labels = labels[range(0, M_TRAINING), :]
X = training_images / 255
y = training_labels
accuracy = _test(thetas, y, X, M_TRAINING)
print 'Trained with {0} examples. Training accuracy: {1}'.format(
M_TRAINING,
accuracy
)
test_labels = labels[range(M_TRAINING, M_TRAINING + M_TESTING), :]
test_images = images[range(M_TRAINING, M_TRAINING + M_TESTING), :]
X = test_images / 255
y = test_labels
accuracy = _test(thetas, y, X, M_TESTING)
print 'Tested with {0} examples. Test accuracy: {1}'.format(M_TESTING,
accuracy)
def train(alpha, n_iter, plot=False):
size_img, rows, cols, images = utils.read_images(
TRAINING_IMAGES_PATH
)
size_lbl, labels = utils.read_labels(TRAINING_LABELS_PATH)
images = images[range(0, M_TRAINING), :]
labels = labels[range(0, M_TRAINING), :]
size_img = M_TRAINING
bias_terms = np.ones([size_img, 1], dtype=np.float64)
images = np.concatenate((bias_terms, images), axis=1).astype(np.float64)
thetas = np.zeros([rows*cols+1, N_LABELS], dtype=np.float64)
costs = np.zeros([n_iter, N_LABELS])
X = images / 255
for i in range(N_LABELS):
# print 'Training a classifier for label {0}'.format(i)
y = np.array([[1 if label == i else 0 for label in labels]]).T
thetas[:, i:i+1], costs[:, i:i+1] = func.gradient_descent(
thetas[:, i:i+1],
y, X, alpha,
n_iter
)
if plot:
plt.plot(costs[:, i:i+1])
plt.show()
return thetas
def process(input_path, destination_path, id_delta=0):
print "processing ", input_path
images = utils.read_images(input_path)
print len(images), " images read"
new_images = utils.convert_all_files(images)
updated_list = [(id + id_delta, image) for id, image in new_images]
print "saving"
utils.save_images_width_id(updated_list, destination_path)
def read_action(action_path):
action_name=get_action_name(action_path)
category=get_category(action_name)
person=get_person(action_name)
all_files=utils.get_files(action_path)
all_files=utils.append_path(action_path+"/",all_files)
frames=utils.read_images(all_files)
return Action(category,person,frames)
def process(input_path, destination_path, id_delta=0):
print "processing ", input_path
images = utils.read_images(input_path)
print len(images), " images read"
new_images = utils.convert_all_files(images)
updated_list = [(id+id_delta, image) for id, image in new_images]
print "saving"
utils.save_images_width_id(updated_list, destination_path)
def train_model(unique_labels, file_dir, num_samples, le_file_path, num_epochs, model_file_path):
model = create_model(unique_labels)
file_list = utils.get_file_list(file_dir, samples=num_samples, training=True)
logger.info('Beginning training model')
images, labels = utils.read_images(file_list, le_file_path, training=True)
model.fit(images, labels, epochs=num_epochs)
logger.info('Finished fitting model')
logger.info(f'Saving model to {model_file_path}')
model.save(model_file_path)
def main():
data = read_df(data_dir/'sentinel_main.csv')
image_list = read_images(image_dir)
for ((state, region), group) in data.groupby(['state', 'region']):
with ThreadPoolExecutor(max_workers=5) as executor:
rows = [(Coordinate(row['lat'], row['lng']), row['cluster'])
for _, row in group.iterrows() if str(row['cluster']) not in image_list]
results = list(tqdm(executor.map(process, rows), desc=f'State: {state}, Region: {region}', total=len(rows)))
def test_processing():
# load images
image_list = utils.read_images('/Users/matti/Documents/forritun/andlit2/')
print len(image_list)
# pre_ids, pre_images = zip(*image_list)
processed_images = utils.convert_all_files(image_list)
print len(processed_images)
# ids, images = zip(*processed_images)
# print " pre: ", pre_ids
# print "post: ", ids
# utils.save_images(images)
utils.save_images_width_id(processed_images)
def test_processing():
# load images
image_list = utils.read_images('/Users/matti/Documents/forritun/andlit2/')
print len(image_list)
# pre_ids, pre_images = zip(*image_list)
processed_images = utils.convert_all_files(image_list)
print len(processed_images)
# ids, images = zip(*processed_images)
# print " pre: ", pre_ids
# print "post: ", ids
# utils.save_images(images)
utils.save_images_width_id(processed_images)
def predict_images(model_file_path, predict_file_dir, le_file_path):
model = tf.keras.models.load_model(model_file_path)
file_list = utils.get_file_list(predict_file_dir)
images, _ = utils.read_images(file_list, le_file_path)
with open(le_file_path, 'rb') as f:
le = pickle.load(f)
predictions = model.predict(images)
predicted_classes = le.inverse_transform(
[np.argmax(prediction) for prediction in predictions])
for idx in range(len(predictions)):
file_name = file_list[idx].split('\\')[-1]
predicted_class = predicted_classes[idx]
logger.info(f'File: {file_name} | Predicted Class: {predicted_class}')
def load_data(self, train_p_data_dir, train_n_data_dir):
p_data = read_images(train_p_data_dir,
self.image_size,
normalize=True,
limit=5000)
n_data = read_images(train_n_data_dir,
self.image_size,
normalize=True,
limit=10000)
random.shuffle(p_data)
random.shuffle(n_data)
# 分割出test集合
split_p = int(len(p_data) * 0.2)
split_n = int(len(n_data) * 0.2)
self.test_p_data = p_data[:split_p]
self.test_n_data = n_data[:split_n]
p_data = p_data[split_p:]
n_data = n_data[split_n:]
split_p = int(len(p_data) * 0.7)
split_n = int(len(n_data) * 0.7)
self.train_p_data = p_data[:split_p]
self.train_n_data = n_data[:split_n]
self.valid_p_data = p_data[split_p:]
self.valid_n_data = n_data[split_n:]
self.train_p_num, self.train_n_num = len(self.train_p_data), len(
self.train_n_data)
self.valid_p_num, self.valid_n_num = len(self.valid_p_data), len(
self.valid_n_data)
# 训练数据自增操作
self.train_p_data += list(map(np.flipud, self.train_p_data))
self.train_n_data += list(map(np.flipud, self.train_n_data))
#self.train_n_data = self.train_n_data[:len(self.train_p_data)]
print("train data size: ", "p:", len(self.train_p_data), "n:",
len(self.train_n_data))
print("valid data size: ", "p:", len(self.valid_p_data), "n:",
len(self.valid_n_data))
print("test data size: ", "p:", len(self.test_p_data), "n:",
len(self.test_n_data))
def load_data(path,batch_size=25):
all_files=utils.get_all_files(path)
all_files=utils.append_path(path,all_files)
images=utils.read_images(all_files)
images=utils.flatten_images(images)
images=map(utils.normalize,images)
images=np.array(images)
n_batches=get_number_of_batches(batch_size,len(images))
def get_batch(i):
return images[i * batch_size: (i+1) * batch_size]
batches=map(get_batch,range(n_batches))
batches = [np.array(batch) for batch in batches]
print("Dataset loaded")
return np.array(batches)
#mean = [0.5, 0.5, 0.5]
#std=[1, 1, 1]
# mean from dataset058:
mean = [0.3823625683879477, 0.3790166856065496, 0.3554138533338805]
std=[0.21754145353302254, 0.21271749678359336, 0.21233947166469555]
#mean = [0.383661700858527, 0.3819784115384924, 0.3588786631614881]
#std=[0.2167717755518767, 0.21201058526724945, 0.21143164036556178]
# Load the dataset
from utils import read_images_stl10 as read_images
from utils import read_labels_stl10 as read_labels
#unlab_set_x = read_images('../data/stl10_binary/unlabeled_X.bin')
test_set_x = read_images('../../data/stl10_binary/test_X.bin')
train_set_x = read_images('../../data/stl10_binary/train_X.bin')
test_set_y = read_labels('../../data/stl10_binary/test_y.bin')
train_set_y = read_labels('../../data/stl10_binary/train_y.bin')
print 'Train set information: '
print (len(train_set_x), type(train_set_x[3]))
print ''
print 'Test set information: '
print (len(test_set_x), type(test_set_x[3]))
print ''
# loading the folds:
with open('../../data/stl10_binary/fold_indices.txt', 'r') as f_folds:
folds = f_folds.readlines()
def test_detection():
image_list = utils.read_images('/Users/matti/Documents/forritun/andlit2/s18')
fd = utils.FaceDetector()
ids, images = zip(*processed_images)
from optim import SGD
from utils import read_images, read_labels
from model import LeNet
parser = argparse.ArgumentParser(
description='Training LeNet on MNIST using NumPy.')
parser.add_argument('data_dir', type=str, help='directory to mnist data')
args = parser.parse_args()
epoch = 10
lr = 0.1
momentum = 0.8
batch = 256
train_data = read_images(os.path.join(args.data_dir,
'train-images.idx3-ubyte'))
train_labels = read_labels(
os.path.join(args.data_dir, 'train-labels.idx1-ubyte'))
test_data = read_images(os.path.join(args.data_dir, 't10k-images.idx3-ubyte'))
test_labels = read_labels(os.path.join(args.data_dir,
't10k-labels.idx1-ubyte'))
# normalize
train_data = (train_data - train_data.mean(
(1, 2), keepdims=True)) / train_data.std((1, 2), keepdims=True)
test_data = (test_data - test_data.mean(
(1, 2), keepdims=True)) / test_data.std((1, 2), keepdims=True)
my_net = LeNet()
optimizer = SGD(my_net.parameters(), lr, momentum)
print "starting...."
# This is where we write the images, if an output_dir is given
# in command line:
out_dir = None
# You'll need at least a path to your image data, please see
# the tutorial coming with this source code on how to prepare
# your image data:
path = '/Users/matti/Documents/forritun/att_faces/'
if len(sys.argv) > 1:
path = sys.argv[1]
print "reading images from " + path
# Now read in the image data. This must be a valid path!
# [X, y] = read_images(path)
input_faces = utils.read_images(path)
# use a random image
# random.seed()
# r = len(X)
# random_index = random.randint(0, r-1)
# print "using image ", random_index, " id: ", y[random_index]
prufu_mynd = None
# test data and label
# prufu_mynd, tl = X[random_index], y[random_index]
# and remove test from the data
# del X[random_index]
# del y[random_index]
def test_detection():
image_list = utils.read_images(
'/Users/matti/Documents/forritun/andlit2/s18')
fd = utils.FaceDetector()
ids, images = zip(*processed_images)
def train(FLAG):
print("Reading dataset...")
# load data
Xtrain, Ytrain = read_images(TRAIN_DIR), read_masks(TRAIN_DIR, onehot=True)
Xtest, Ytest = read_images(VAL_DIR), read_masks(VAL_DIR, onehot=True)
track = [
"hw3-train-validation/validation/0008",
"hw3-train-validation/validation/0097",
"hw3-train-validation/validation/0107"
]
Xtrack, Ytrack = read_list(track)
vgg16 = VGG16(classes=7, shape=(256, 256, 3))
vgg16.build(vgg16_npy_path=FLAG.init_from,
mode=FLAG.mode,
keep_prob=FLAG.keep_prob)
saver = tf.train.Saver(tf.global_variables(), max_to_keep=1)
checkpoint_path = os.path.join(FLAG.save_dir, 'model.ckpt')
def initialize_uninitialized(sess):
global_vars = tf.global_variables()
is_not_initialized = sess.run(
[tf.is_variable_initialized(var) for var in global_vars])
not_initialized_vars = [
v for (v, f) in zip(global_vars, is_not_initialized) if not f
]
if len(not_initialized_vars):
sess.run(tf.variables_initializer(not_initialized_vars))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
# hyper parameters
batch_size = 32
epoch = 500
early_stop_patience = 50
min_delta = 0.0001
opt_type = 'adam'
# recorder
epoch_counter = 0
# optimizer
global_step = tf.Variable(0, trainable=False)
# Passing global_step to minimize() will increment it at each step.
if opt_type is 'sgd':
start_learning_rate = FLAG.lr
half_cycle = 2000
learning_rate = tf.train.exponential_decay(start_learning_rate,
global_step,
half_cycle,
0.5,
staircase=True)
opt = tf.train.MomentumOptimizer(learning_rate=learning_rate,
momentum=0.9,
use_nesterov=True)
else:
start_learning_rate = FLAG.lr
half_cycle = 2000
learning_rate = tf.train.exponential_decay(start_learning_rate,
global_step,
half_cycle,
0.5,
staircase=True)
opt = tf.train.AdamOptimizer(learning_rate=learning_rate)
obj = vgg16.loss
train_op = opt.minimize(obj, global_step=global_step)
# progress bar
ptrain = IntProgress()
pval = IntProgress()
display(ptrain)
display(pval)
ptrain.max = int(Xtrain.shape[0] / batch_size)
pval.max = int(Xtest.shape[0] / batch_size)
# re-initialize
initialize_uninitialized(sess)
# reset due to adding a new task
patience_counter = 0
current_best_val_loss = np.float('Inf')
# optimize when the aggregated obj
while (patience_counter < early_stop_patience
and epoch_counter < epoch):
# start training
stime = time.time()
bar_train = Bar(
'Training',
max=int(Xtrain.shape[0] / batch_size),
suffix='%(index)d/%(max)d - %(percent).1f%% - %(eta)ds')
bar_val = Bar(
'Validation',
max=int(Xtest.shape[0] / batch_size),
suffix='%(index)d/%(max)d - %(percent).1f%% - %(eta)ds')
train_loss, train_accu = 0.0, 0.0
for i in range(int(Xtrain.shape[0] / batch_size)):
st = i * batch_size
ed = (i + 1) * batch_size
loss, accu, _ = sess.run(
[obj, vgg16.accuracy, train_op],
feed_dict={
vgg16.x: Xtrain[st:ed, :],
vgg16.y: Ytrain[st:ed, :],
vgg16.is_train: True
})
train_loss += loss
train_accu += accu
ptrain.value += 1
ptrain.description = "Training %s/%s" % (ptrain.value,
ptrain.max)
train_loss = train_loss / ptrain.value
train_accu = train_accu / ptrain.value
# validation
val_loss = 0
val_accu = 0
for i in range(int(Xtest.shape[0] / batch_size)):
st = i * batch_size
ed = (i + 1) * batch_size
loss, accu = sess.run(
[obj, vgg16.accuracy],
feed_dict={
vgg16.x: Xtest[st:ed, :],
vgg16.y: Ytest[st:ed, :],
vgg16.is_train: False
})
val_loss += loss
val_accu += accu
pval.value += 1
pval.description = "Testing %s/%s" % (pval.value, pval.value)
val_loss = val_loss / pval.value
val_accu = val_accu / pval.value
# plot
if epoch_counter % 10 == 0:
Xplot = sess.run(vgg16.pred,
feed_dict={
vgg16.x: Xtrack[:, :],
vgg16.y: Ytrack[:, :],
vgg16.is_train: False
})
for i, fname in enumerate(track):
saveimg = skimage.transform.resize(Xplot[i],
output_shape=(512, 512),
order=0,
preserve_range=True,
clip=False)
saveimg = label2rgb(saveimg)
imageio.imwrite(
os.path.join(
FLAG.save_dir,
os.path.basename(fname) + "_pred_" +
str(epoch_counter) + ".png"), saveimg)
print(
os.path.join(
FLAG.save_dir,
os.path.basename(fname) + "_pred_" +
str(epoch_counter) + ".png"))
# early stopping check
if (current_best_val_loss - val_loss) > min_delta:
current_best_val_loss = val_loss
patience_counter = 0
saver.save(sess, checkpoint_path, global_step=epoch_counter)
print("save in %s" % checkpoint_path)
else:
patience_counter += 1
# shuffle Xtrain and Ytrain in the next epoch
idx = np.random.permutation(Xtrain.shape[0])
Xtrain, Ytrain = Xtrain[idx, :, :, :], Ytrain[idx, :]
# epoch end
epoch_counter += 1
ptrain.value = 0
pval.value = 0
bar_train.finish()
bar_val.finish()
print(
"Epoch %s (%s), %s sec >> train loss: %.4f, train accu: %.4f, val loss: %.4f, val accu: %.4f"
% (epoch_counter, patience_counter,
round(time.time() - stime,
2), train_loss, train_accu, val_loss, val_accu))
t_vars = tf.trainable_variables()
slim.model_analyzer.analyze_vars(t_vars, print_info=True)
with tf.control_dependencies(
tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='discriminator')):
train_D = tf.train.AdamOptimizer(0.0002,
beta1=0.5).minimize(model.loss_D,
var_list=model.vars_D)
with tf.control_dependencies(
tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='generator')):
train_G = tf.train.AdamOptimizer(0.0002, beta1=0.5).minimize(
model.loss_G, global_step=model.global_step, var_list=model.vars_G)
sess_config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
images, labels = read_images("data", "folder")
num_iters = len(images) // config.BATCH_SIZE
cnt = 0
length = 5
sample_noise = np.random.uniform(
-1., 1., size=[length * length, 1, 1, config.LATENT_DIM])
with tf.Session(config=sess_config) as sess:
saver = tf.train.Saver(tf.global_variables(), max_to_keep=1000)
#summary_op = tf.summary.merge_all()
init = tf.global_variables_initializer()
sess.run(init)
model_checkpoint_name = config.PATH_CHECKPOINT + "/model.ckpt"
if config.IS_CONTINUE:
else:
image_patch = img[i:(i + template_h), j:(j + template_w)]
image_template = template
results[method][i, j] = value['func'](image_patch, image_template)
# for each result get max(or min) and plot corresponding matching result
for method, res in results.items():
im_copy = img.copy()
if method != 'Normalized Cross-Correlation':
top_left = np.unravel_index(res.argmin(), res.shape)[::-1]
else:
top_left = np.unravel_index(res.argmax(), res.shape)[::-1]
bottom_right = (top_left[0] + template_w, top_left[1] + template_h)
cv2.rectangle(im_copy, top_left, bottom_right, 255, 2)
print_matching_result(im_copy, res, method)
if __name__ == '__main__':
image, template = read_images('0001.jpg', '0001_template.jpg', DATA_DIR)
opencv_methods = ['cv2.TM_CCORR_NORMED', 'cv2.TM_SQDIFF']
custom_methods = {
'Sum of Absolute Distances': {'func': sad, 'normalized': False},
'Normalized Sum of Absolute Distances': {'func': sad, 'normalized': True},
'Sum of Square Distances': {'func': ssd, 'normalized': False},
'Normalized Cross-Correlation': {'func': ncc, 'normalized': True}
}
# template_matching_opencv(image, template, opencv_methods)
template_matching_numpy(image, template, custom_methods)
requiredArgs = parser.add_argument_group('required arguments')
requiredArgs.add_argument(
"-i",
"--input",
help="Folder containing an images/ folder and Calibration.xml",
required=True)
parser.add_argument(
"--rerun",
action='store_true',
help=
"Use this flag if this is the second run of the script, first run stores the point grid visible which is used from that point on"
)
args = parser.parse_args()
calibrations = read_calibration(os.path.join(args.input, "Calibration.xml"))
images = read_images(os.path.join(args.input, "images"))
if not args.rerun:
grid = generate_grid()
grid = filter_grid(grid, calibrations)
np.save('grid.npy', grid)
else:
grid = np.load('grid.npy')
draw_points_on_images(grid, calibrations, images)
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
colors = [
'#808080', '#000000', '#FF0000', '#800000', '#808000', '#008000',
'#00FFFF', '#008080', '#800080'
]
for i, calibration in enumerate(calibrations):
"""
SAA - self adaptive algorithm
FAS - fast adaptive similarity filter
Images from https://pixabay.com/images/search/
"""
import time
import utils
import constants
import NEAVF
start = time.time()
# Read, resize and plot images
size = (300, 400)
img_list = utils.read_images(constants.DATASET_PATH)
img_list = utils.resize_img(img_list, size)
# utils.show_image_list(img_list, "Initial images!")
img_list_gauss = []
img_list_impulsive = []
index = 0
for img in img_list:
img_list_gauss.append(utils.gaussian_noise(img, 0.1))
img_list_impulsive.append(utils.salt_and_pepper(img, 0.1))
index += 1
# utils.show_image_list(img_list_gauss, "Images with gaussian noise! 0.1")
# utils.show_image_list(img_list_impulsive, "Image with salt and pepper noise! 0.1")
# Two images with 2 values of noise
img_list_gauss_005 = []
config.MODE = "train"
model = GANnomaly(config)
t_vars = tf.trainable_variables()
slim.model_analyzer.analyze_vars(t_vars, print_info=True)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='discriminator')):
train_D = tf.train.AdamOptimizer(0.0002, beta1=0.5).minimize(model.loss_D, var_list=model.vars_D)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='generator') + tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='encoder')):
train_G = tf.train.AdamOptimizer(0.0002, beta1=0.5).minimize(model.loss_G, global_step=model.global_step, var_list=model.vars_G)
sess_config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
images, labels = read_images(config.PATH_DATA, "folder")
num_iters = len(images) // config.BATCH_SIZE
test_images, test_labels = read_images(config.PATH_TEST, "folder")
test_num_iters = len(test_images) // config.BATCH_SIZE
cnt = 0
length = 6
best_auc = 0
scores_out = []
labels_out = []
with tf.Session(config=sess_config) as sess:
saver = tf.train.Saver(tf.global_variables(), max_to_keep=10000)
init = tf.global_variables_initializer()
def run_experiment(params, dropout, bn):
tf.reset_default_graph()
figname_suffix = "a_"
dropout_value = 1.0
if dropout:
dropout_value = params['dropout']
figname_suffix += "dropout"
if bn:
figname_suffix += "bn"
train_step, \
cost, \
accuracy, \
y_pred, \
y_pred_cls, \
y_true_cls, \
placeholders = create_network(
params['img_size'],
params['num_channels'],
params['num_classes'],
params['shape1'],
params['shape2'],
params['num_fc_layer1_output'],
params['num_fc_layer2_output'],
params['learning_rate'],
bn
)
saver = tf.train.Saver()
if not os.path.exists(params['save_dir']):
os.makedirs(params['save_dir'])
(train_acc, cost,
test_acc) = train_network(params['data'], train_step, cost, accuracy,
params['num_iterations'],
params['train_batch_size'], dropout_value,
placeholders, saver, params['save_dir'],
params['plot_dir'], params['log_dir'],
params['display_step'], figname_suffix)
cls_true, cls_pred, acc, y_pred_probs, x_all, y_all\
= test_network(params['test_batch_size'],
placeholders,
1.0,
saver,
params['save_dir'],
accuracy,
y_pred,
y_pred_cls,
y_true_cls,
params['data'])
plot_confusion_matrix(cls_true=cls_true,
cls_pred=cls_pred,
output_dir=plot_dir,
fig_name="confusion_matrix" + "_" + figname_suffix)
plot_images(output_dir=plot_dir,
fig_name="images" + "_" + figname_suffix,
img_shape=(params['img_size'], params['img_size'],
params['num_channels']),
images=x_all,
cls_true=y_all,
cls_pred=cls_pred,
prob_pred=y_pred_probs,
logits=None,
y_pred=None)
y_all_label = list(map(lambda x: np.argmax(x), y_all))
chosen = list(
filter(lambda x: x[0] == params['real_label'], zip(y_all_label,
x_all)))
chosen_class = np.zeros(10)
chosen_class[params['chosen_label']] = 1
image_index = np.random.randint(len(chosen) - 1)
x_image_list = create_plot_adversarial_images(
x=placeholders['x'],
keep_prob=placeholders['fc_layer1_keep_prob'],
phase_train=placeholders['phase_train'],
img_shape=(params['img_size'], params['img_size'],
params['num_channels']),
x_image=chosen[image_index][1],
y_conv=y_pred,
y_label=chosen_class,
lr=0.05,
n_steps=12,
saver=saver,
output_dir=params['plot_dir'],
save_dir=params['save_dir'],
fig_name=str(chosen_label) + "_" + str(real_label),
)
# create a second network, train and test both
train_step2, \
cost2, \
accuracy2, \
y_pred2, \
y_pred_cls2, \
y_true_cls2, \
placeholders2 = create_network2(
params['img_size'],
params['num_channels'],
params['num_classes'],
params['shape1'],
params['shape2'],
params['num_fc_layer1_output'],
params['num_fc_layer2_output'],
params['learning_rate'],
bn
)
(train_acc2, cost2,
test_acc2) = train_network(params['data'], train_step2, cost2, accuracy2,
params['num_iterations'],
params['train_batch_size'], dropout_value,
placeholders2, saver, params['save_dir2'],
params['plot_dir'], params['log_dir'],
params['display_step'], figname_suffix)
if params['dataset'] == 'mnist':
gray = True
else:
gray = False
image_list, img_shape = read_images(
params['plot_dir'] + str(chosen_label) + "_" + str(real_label) + "/",
gray)
xx_image_list = []
for img in x_image_list:
img = img.flatten().reshape(784)
xx_image_list += [img]
result1 = test_adversial_images(y_pred, saver, params['save_dir'],
placeholders['x'],
placeholders['fc_layer1_keep_prob'],
placeholders['phase_train'], xx_image_list,
img_shape)
result2 = test_adversial_images(y_pred2, saver, params['save_dir2'],
placeholders2['x'],
placeholders2['fc_layer1_keep_prob'],
placeholders2['phase_train'],
xx_image_list, img_shape)
fig = plt.figure(figsize=(25, 35))
plt.clf()
rows = 6
cols = 4
for i in range(1, len(result1)):
ax = fig.add_subplot(rows, cols, i)
ax.imshow(result1[i][2], cmap='binary')
ax.set_title('Label1: {0} Prob1: {1}%'.format(result1[i][0],
result1[i][1]))
ax = fig.add_subplot(rows, cols, i + 1)
ax.imshow(result2[i][2], cmap='binary')
ax.set_title('Label2: {0} Prob2: {1}%'.format(result2[i][0],
result2[i][1]))
i += 2
plt.savefig(params['plot_dir'] + str(chosen_label) + "_" +
str(real_label) + "/compare.png")
return train_acc, cost, test_acc, figname_suffix, acc
config = Config()
model = DCGAN(config)
config.BATCH_SIZE = 1
t_vars = tf.trainable_variables()
slim.model_analyzer.analyze_vars(t_vars, print_info=True)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='discriminator')):
train_D = tf.train.AdamOptimizer(0.0002, beta1=0.5).minimize(model.loss_D, var_list=model.vars_D)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS, scope='generator')):
train_G = tf.train.AdamOptimizer(0.0002, beta1=0.5).minimize(model.loss_G, global_step=model.global_step, var_list=model.vars_G)
sess_config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
images, labels = read_images(config.PATH_DATA, "folder")
num_iters = len(images) // config.BATCH_SIZE
cnt = 0
length = 5
sample_noise = np.random.uniform(-1., 1., size=[length*length, 1, 1, config.LATENT_DIM])
img_path = "temp"
if not os.path.isdir("temp"):
os.mkdir("temp")
with tf.Session(config=sess_config) as sess:
saver = tf.train.Saver(tf.global_variables(), max_to_keep=1000)
#summary_op = tf.summary.merge_all()
init = tf.global_variables_initializer()
sess.run(init)
# set up detectors
detector = BodyDetector(cascade_fn="../data/people/body10/haarcascade_fullbody.xml", scaleFactor=1.05,
minNeighbors=1)
# detector = BodyDetector(cascade_fn="../data/people/body10/haarcascade_upperbody.xml")
# detector = BodyDetector(cascade_fn="../data/people/body10/haarcascade_lowerbody.xml")
# detector = BodyDetector(cascade_fn="/Users/Utilizador/opencv-3.0.0/data/haarcascades/haarcascade_fullbody.xml", scaleFactor=1.05,
# minNeighbors=1)
# detector = BodyDetector(cascade_fn="/Users/Utilizador/opencv-3.0.0/data/haarcascades/haarcascade_upperbody.xml")
# detector = BodyDetector(cascade_fn="/Users/Utilizador/opencv-3.0.0/data/haarcascades/haarcascade_lowerbody.xml")
# rect = detector.detect(imgOut)
[Names, X, y] = read_images("../data/people/INRIAPerson/Test/pos",sz=None)
# [Names, X, y] = sorted(([Names, X, y]), key=lambda image:X[0])
imgBGR = []
#for all images in list X
for imgName, imgOut in zip(Names,X):
# detection
for i,r in enumerate(detector.detect(imgOut)):
x0,y0,x1,y1 = r
imgBGR = cv2.cvtColor(imgOut,cv2.COLOR_GRAY2BGR)
cv2.rectangle(imgBGR, (x0,y0),(x1,y1),(0,255,0),1)
# display image or write to file
cv2.imshow("People detected: " + imgName, imgBGR)
print "starting...."
# This is where we write the images, if an output_dir is given
# in command line:
out_dir = None
# You'll need at least a path to your image data, please see
# the tutorial coming with this source code on how to prepare
# your image data:
path = '/Users/matti/Documents/forritun/att_faces/'
if len(sys.argv) > 1:
path = sys.argv[1]
print "reading images from " + path
# Now read in the image data. This must be a valid path!
# [X, y] = read_images(path)
input_faces = utils.read_images(path)
# use a random image
# random.seed()
# r = len(X)
# random_index = random.randint(0, r-1)
# print "using image ", random_index, " id: ", y[random_index]
prufu_mynd = None
# test data and label
# prufu_mynd, tl = X[random_index], y[random_index]
# and remove test from the data
# del X[random_index]
# del y[random_index]
id = '#1'
TAG = '_6class_T2'
DIM = '10x10'
CONSTRUCTION_METHOD = 'stacked' # 'image' or 'matrix' or 'stacked'
CSV_FILEPATH = 'data/test_labels' + TAG + '.txt'
DATA_FILEPATH = 'data/test/complexbaseband/' + DIM + TAG + '/'
NUMBER_OF_CLASSES = 6
WEIGHTS = 'CNN/models/weights_' + DIM + '_' + id + '.best.hdf5'
MODEL = 'CNN/models/model_' + DIM + '_' + id + '.json'
object_classes = ['noObject', 'circle', 'square', 'rect', 'Lrect', 'triangle']
train_df, input_shape = data_from_csv_nonimage(CSV_FILEPATH, DATA_FILEPATH)
if CONSTRUCTION_METHOD == 'image':
train_df, input_shape = data_from_csv(CSV_FILEPATH, DATA_FILEPATH)
x_test_channelb = read_images(train_df.channelb.values, input_shape)
x_test_channelc = read_images(train_df.channelc.values, input_shape)
x_test_channeld = read_images(train_df.channeld.values, input_shape)
elif CONSTRUCTION_METHOD == 'matrix':
train_df, input_shape = data_from_csv_nonimage(CSV_FILEPATH, DATA_FILEPATH)
x_test_channelb = read_data(train_df.channelb.values, input_shape)
x_test_channelc = read_data(train_df.channelc.values, input_shape)
x_test_channeld = read_data(train_df.channeld.values, input_shape)
else:
train_df, input_shape = data_from_csv_nonimage_6channel(
CSV_FILEPATH, DATA_FILEPATH)
x_test_channelbcd = read_data(train_df.channelbcd.values, input_shape)
labels = train_df.object.values
labels = keras.utils.to_categorical(labels, NUMBER_OF_CLASSES)
import utils from neuralnetwork import NeuralNetworkModel import autograd.numpy as np import pickle import matplotlib.pyplot as plt from utils import plt_image TRAINING_IMAGES_PATH = 'mnistset/train-images-idx3-ubyte.gz' TRAINING_LABELS_PATH = 'mnistset/train-labels-idx1-ubyte.gz' TESTING_IMAGES_PATH = 'mnistset/t10k-images-idx3-ubyte.gz' TESTING_LABELS_PATH = 'mnistset/t10k-labels-idx1-ubyte.gz' MODELS_PATH = 'mnistmodels/' print 'Reading images from %s' % TRAINING_IMAGES_PATH images = utils.read_images(TRAINING_IMAGES_PATH) print 'Reading labels from %s' % TRAINING_LABELS_PATH labels = utils.read_labels(TRAINING_LABELS_PATH) print 'Reading images from %s' % TESTING_IMAGES_PATH testing_images = utils.read_images(TESTING_IMAGES_PATH) print 'Reading labels from %s' % TESTING_LABELS_PATH testing_labels = utils.read_labels(TESTING_LABELS_PATH) training_images = images[5000:-5000] training_labels = labels[5000:-5000] validating_images = np.concatenate([images[:5000], images[-5000:]]) validating_labels = np.concatenate([labels[:5000], labels[-5000:]]) np.set_printoptions(suppress=True) nnm = None
import utils from neuralnetwork import NeuralNetworkModel import autograd.numpy as np import pickle import matplotlib.pyplot as plt from utils import plt_image TRAINING_IMAGES_PATH = 'mnistset/train-images-idx3-ubyte.gz' TRAINING_LABELS_PATH = 'mnistset/train-labels-idx1-ubyte.gz' TESTING_IMAGES_PATH = 'mnistset/t10k-images-idx3-ubyte.gz' TESTING_LABELS_PATH = 'mnistset/t10k-labels-idx1-ubyte.gz' MODELS_PATH = 'mnistmodels/' print 'Reading images from %s' % TRAINING_IMAGES_PATH images = utils.read_images(TRAINING_IMAGES_PATH) print 'Reading labels from %s' % TRAINING_LABELS_PATH labels = utils.read_labels(TRAINING_LABELS_PATH) print 'Reading images from %s' % TESTING_IMAGES_PATH testing_images = utils.read_images(TESTING_IMAGES_PATH) print 'Reading labels from %s' % TESTING_LABELS_PATH testing_labels = utils.read_labels(TESTING_LABELS_PATH) training_images = images[5000:-5000] training_labels = labels[5000:-5000] validating_images = np.concatenate([images[:5000], images[-5000:]]) validating_labels = np.concatenate([labels[:5000], labels[-5000:]]) np.set_printoptions(suppress=True) nnm = None
discriminator_costs.append(discriminator_cost)
display.clear_output(wait=True)
print('Epoch: {}'.format(epoch+1))
print('Generator loss: {}'.format(generator_loss))
print('Discriminator loss: {}'.format(discriminator_loss))
noise = tf.random.normal([9, noise_size])
generate_print_save(generator, noise, epoch)
# plot the learning curves of the generator and discriminator
plt.plot(np.squeeze(generator_costs))
plt.plot(np.squeeze(discriminator_costs))
plt.show()
# download images
train_set = read_images(data_dir='data/*.png')
# let us look at few images from the dataset
print_images(images=train_set)
# setup number of channels, noise size, learning rate and beta_1
channels = 3
noise_size = 100
learning_rate = 0.0001
beta_1 = 0.5
# create an instance of generator and discriminator, their optimizers and checkpoints
generator, generator_optimizer, generator_checkpoint = create_generator(channels, noise_size, learning_rate, beta_1)
discriminator, discriminator_optimizer, discriminator_checkpoint = create_discriminator(channels, learning_rate, beta_1)
# train generator and discriminator