def load_cube(self, image, padding_strategy, padded_size=None):
output_array = None
if padded_size is None:
padded_size = self.roi_size
self.padded_size = padded_size
diagonal_length = np.sqrt(np.sum(np.power(padded_size, 2)))
diagonal_size = [diagonal_length , diagonal_length , diagonal_length]
for idx, i in enumerate(self.roi_size):
if (diagonal_size[idx] - i) % 2 != 0:
diagonal_size[idx] += 1
pass
# self.image_z_spacing = image.GetSpacing()[2]
diagonal_size = np.array(diagonal_size)
if padding_strategy == 'neighbour':
max_sz = diagonal_size
max_start = np.int16(np.floor(self.roi_center - max_sz / 2))
max_end = np.int16(np.floor(self.roi_center + max_sz / 2))
output_array = np.float32(ImageHelper.get_roi(image, max_start.tolist(), np.int16(max_sz).tolist()))
"""
output_array = np.float32(ndarray[
max_start[0]:max_end[0],
max_start[1]:max_end[1],
max_start[2]:max_end[2]
])
"""
elif padding_strategy in [None, 'zero', 'repeat']:
output_array = np.float32(ImageHelper.get_roi(image, self.roi_start_idx.tolist(), self.roi_idx_size.tolist()))
output_array = ImageHelper.pad_image(output_array, diagonal_size, mode=padding_strategy)
"""
output_array = np.float32(ndarray[
self.roi_start_idx[0]:self.roi_end_idx[0],
self.roi_start_idx[1]:self.roi_end_idx[1],
self.roi_start_idx[2]:self.roi_end_idx[2]
])
"""
else:
raise ValueError("Unrecognized padding mode {0}".format(padding_strategy))
self.cube = output_array
# self.test(image)
return self.cube
def infer(self, sess):
# 加载模型
ckpt = tf.train.get_checkpoint_state(self.checkpoint_dir)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
self.saver.restore(sess, ckpt.model_checkpoint_path)
image_helper = ImageHelper()
fake_imgs = sess.run(
self.fake_imgs,
feed_dict={self.rand_noises: np.random.normal(size=[self.batch_size, self.noise_dim])})
img_name = "{}/infer-image".format(self.hparams.sample_dir)
image_helper.save_imgs(fake_imgs,
img_name=img_name)
tf.logging.info("====== generate images in file: {} ======".format(img_name))
def __init__(self, image_filename, model_object, input_cube_size=None):
self.shutdown_signal = False
self.worker_threads = []
self.worker_thread_count = 1
self.requires_padding = False
self.model_cube_size = np.array([15, 15, 23])
if input_cube_size is None:
self.input_cube_size = self.model_cube_size
else:
self.input_cube_size = np.array(input_cube_size)
for i in range(3):
if self.input_cube_size[i] != self.model_cube_size[i]:
self.requires_padding = True
if self.input_cube_size[i] > self.model_cube_size[i]:
raise ValueError(
"Input Cube size cannot be larger than model cube size " +
str(self.model_cube_size))
self.q = None
self.spacing_factor, self.image = ImageHelper.read_image(
image_filename, correct_spacing=True)
self.org_ndarray = sitk.GetArrayFromImage(self.image)
ndarray = np.float32(np.swapaxes(self.org_ndarray, 0, 2))
ndarray -= np.mean(ndarray, axis=0)
ndarray /= np.std(ndarray, axis=0)
self.ndarray = ndarray
self.model = model_object
def worker(self):
while not self.shutdown_signal:
q = self.q
cube_size = self.input_cube_size
cube_half_size = cube_size / 2
image_size = list(self.ndarray.shape)
x_range = [[
floor(int(image_size[0] * .15) + cube_half_size[0]),
floor(int(image_size[0] * .45) - cube_half_size[0])
],
[
floor(int(image_size[0] * .55) + cube_half_size[0]),
floor(int(image_size[0] * .80) - cube_half_size[0])
]]
y_range = [
floor(int(image_size[1] * .25) + cube_half_size[1]),
floor(int(image_size[1] * .75) - cube_half_size[1])
]
z_range = [
floor(int(image_size[2] * .3) + cube_half_size[2]),
floor(int(image_size[2] * .6) - cube_half_size[2])
]
random_center = [
random.randint(*x_range[random.randint(0, 1)]),
random.randint(*y_range),
random.randint(*z_range)
]
cube = self.ndarray[
floor(random_center[0] -
cube_half_size[0]):floor(random_center[0] +
cube_half_size[0]),
floor(random_center[1] -
cube_half_size[1]):floor(random_center[1] +
cube_half_size[1]),
floor(random_center[2] -
cube_half_size[2]):floor(random_center[2] +
cube_half_size[2]), ]
cube = ImageHelper.pad_image(cube, self.model_cube_size)
cube = ImageHelper.add_dim(cube)
result = self.model.test({'image': np.array([cube])})
q.put([random_center, cube, result])
def __init__(self, camera, output_path, app):
self.output_path = output_path
self.last_file_path = None
self.app = app
self.camera = camera
self.camera_overlay = CameraOverlay(self.camera)
self.printer = Printer()
self.green_btn = Button(GREEN_BTN_PIN)
self.red_btn = Button(RED_BTN_PIN)
self.green_btn.when_pressed = self.pressed_capture_button
self.red_btn.when_pressed = self.pressed_reject_print_button
self.ui = UI(self.camera_overlay, camera.picam.resolution)
self.image_helper = ImageHelper(camera.picam.resolution)
self.rotary = Rotary(ROTARY_CLK_PIN, ROTARY_DT_PIN, upperBound=5)
self.camera.show_preview(True)
self.ui.show_main_screen()
self.waiting_for_confirm = False
self.busy = False
def load_cubes_upscale(self):
cube_shape = None
counter = 0
new_rois = []
for idx in self.image_data:
if len(self.image_data[idx]) == 0:
continue
roi_list = self.image_data[idx]
image_path = os.path.abspath(self.input_path + idx + '.img.gz')
image, rois = ImageHelper.scale_rois(
image_path,
roi_list,
0.5,
interpolation_strategy=self.interpolation_strategy)
new_rois.extend(rois)
self.calc_bounds(new_rois)
scaled_rois = []
for idx in self.image_data:
if len(self.image_data[idx]) == 0:
continue
sys.stdout.write("\rLoading Images {0} out of {1}...".format(
counter, len(self.image_data)))
sys.stdout.flush()
counter += 1
roi_list = self.image_data[idx]
image_path = os.path.abspath(self.input_path + idx + '.img.gz')
image, rois = ImageHelper.scale_rois(
image_path,
roi_list,
0.5,
interpolation_strategy=self.interpolation_strategy)
cube_shape = self.load_image_cubes(image, rois, self.roi_size_max)
scaled_rois.extend(rois)
return scaled_rois, self.roi_size_max
def write_post(self, item):
filename = '{0}/{1}.md'.format(self.posts_dir, get(item, 'post_name'))
img = ImageHelper(
images_dir='static/images/{0}'.format(get(item, 'post_name')),
images_url_root='/images/{0}'.format(get(item, 'post_name'))
)
self.check_posts_dir()
title = get(item, 'title')
published = datetime.strptime(get(item, 'post_date'), '%Y-%m-%d %H:%M:%S').isoformat()
status = get(item, 'status')
content = PostConverter(get(item, 'content:encoded')) \
.convert_wordpress_tags() \
.convert_images_and_galleries() \
.convert_html_elements() \
.convert_youtube() \
.clean_hyperlinks() \
.convert_code_blocks() \
.strip_excessive_newlines() \
.to_string()
first_image = img.convert_image(img.find_first_image(content))
for url in img.find_all_images(content):
image_url = img.convert_image(url)
content = content.replace(url, image_url)
print("\n\nCONVERTING: {0}\n\nFilename: {1}\nDate: {2}\nStatus: {3}".format(title, filename, published, status))
with open(filename, 'w') as f:
self.write_header(f, first_image, published, title)
f.write(content)
def get_image(self, id):
is_last_run = False
self.image_locks[id].acquire()
if id not in self.images:
image_path = os.path.abspath(self.input_path + id + '.img.gz')
self.images[id] = ImageHelper.read_image(image_path)
self.image_locks[id].release()
return self.images[id], is_last_run
def load_cubes_raw(self):
cube_shape = None
counter = 0
for idx in self.image_data:
sys.stdout.write("\rLoading Images {0} out of {1}...".format(
counter, len(self.image_data)))
sys.stdout.flush()
counter += 1
image_path = os.path.abspath(self.input_path + idx + '.img.gz')
image = ImageHelper.read_image(image_path, correct_spacing=False)
roi_list = self.image_data[idx]
cube_shape = self.load_image_cubes(image, roi_list,
self.roi_size_max)
return self.roi_size_max
def extract_cubes_one_image(self):
counter = 0
rois = []
self.read_rois()
idx = None
idx = '081832_20060531_1149725568.2203'
roi_list = self.image_data[idx]
spacing_factor = 1
if idx not in self.images:
image_path = os.path.abspath(self.input_path + idx + '.img.gz')
# spacing_factor, self.images[idx] = ImageHelper.read_image(image_path, correct_spacing=True)
spacing_factor = 1
self.images[idx] = ImageHelper.read_image(image_path,
correct_spacing=False)
the_one_roi = None
for roi in roi_list:
if roi.roi_start_idx[0] != 101:
continue
the_one_roi = roi
roi.rescale_depth(spacing_factor)
cube = roi.load_cube(self.images[idx], padded_size=self.cube_size)
assert (list(cube.shape) == self.cube_max_size)
counter += 1
print(counter)
counter = 0
roi_list = [the_one_roi]
for roi_1 in roi_list:
counter += 1
roi_1.lookup_table = self.class_lookup
roi_1.y = self.class_lookup[roi_1.roi_class]
for roi_2 in roi_list:
roi_1.intersects(roi_2)
print(counter)
rois.extend(self.image_data[idx])
pickle.dump([the_one_roi], open("one_image.p", "wb"))
def get_cube(self, rotation=None, padding=True, as_nd=True):
cube_shape = np.array(self.cube.shape)
half_cube_shape = cube_shape/2
half_roi_size = (self.roi_idx_size/2)
idx_start = np.int16(np.floor(half_cube_shape - half_roi_size))
idx_end = np.int16(np.floor(half_cube_shape + half_roi_size))
actual_size = self.roi_idx_size
padded_size = self.padded_size
# print(cube_shape)
# print(padded_size)
# print(actual_size)
if rotation is not None:
if sum(rotation) == 0:
rotation = None
ndarray = None
if rotation is None:
# No rotation, skip all these cpu intensive operations
ndarray = self.cube[
idx_start[0]:idx_end[0],
idx_start[1]:idx_end[1],
idx_start[2]:idx_end[2]
]
else:
# We have rotation
rotation = np.float32(rotation) * (math.pi/180)
rotation_center = half_cube_shape
rigid_euler = sitk.Euler3DTransform()
# rigid_euler.SetRotation(float(rotation[0]), float(rotation[1]), float(rotation[2]))
rigid_euler.SetRotation(float(rotation[0]), float(rotation[1]), float(rotation[2]))
rigid_euler.SetCenter(rotation_center - 0.5)
ndarray = np.swapaxes(
sitk.GetArrayFromImage(
sitk.Resample(
sitk.GetImageFromArray(np.swapaxes(self.cube, 0, 2)),
rigid_euler,
sitk.sitkLinear,
0,
sitk.sitkFloat32
)
),
0, 2)
new_center = (np.array(ndarray.shape) / 2)
new_start = np.int16(new_center - half_roi_size)
new_end = np.int16(new_center + half_roi_size)
ndarray = np.float32(ndarray[
new_start[0]:new_end[0],
new_start[1]:new_end[1],
new_start[2]:new_end[2]
])
# Padding is done after rotation, maybe try to pad before rotation
if padding:
# print(ndarray.shape)
# print(padded_size)
ndarray = ImageHelper.pad_image(ndarray, padded_size, mode='zero')
# print(ndarray.shape)
return ndarray if as_nd else sitk.GetImageFromArray(np.swapaxes(ndarray, 0, 2))
self.img_helper.save_image(generated, epoch, "generated/wgan/",
self.dataset)
def predict_noise(self, size):
noise = np.random.normal(0, 1, (size, self.latent_dim))
return self.generator.predict(noise)
def plot_loss(self, history):
hist = pd.DataFrame(history)
plt.figure(figsize=(20, 5))
for col in hist.columns:
plt.plot(hist[col], label=col)
plt.legend()
plt.ylabel("loss")
plt.xlabel("epochs")
plt.show()
if __name__ == "__main__":
from image_helper import ImageHelper
(X, _), (_, _) = mnist.load_data()
X_train = X / 127.5 - 1.
X_train = np.expand_dims(X_train, axis=3)
wgan = WGAN(img_shape=X_train[0].shape,
latent_dim=100,
img_helper=ImageHelper())
wgan.train(epochs=40000,
train_data=X_train,
batch_size=128,
sample_interval=100)
np.random.seed(1)
time_start_load_everything = time.time()
parser = argparse.ArgumentParser(description='PPDL')
parser.add_argument('--params', dest='params')
args = parser.parse_args()
with open(f'./{args.params}', 'r') as f:
params_loaded = yaml.load(f)
current_time = datetime.datetime.now().strftime('%b.%d_%H.%M.%S')
if params_loaded['type'] == config.TYPE_LOAN:
helper = LoanHelper(current_time=current_time,
params=params_loaded,
name=params_loaded.get('name', 'loan'))
helper.load_data(params_loaded)
elif params_loaded['type'] == config.TYPE_CIFAR:
helper = ImageHelper(current_time=current_time,
params=params_loaded,
name=params_loaded.get('name', 'cifar'))
helper.load_data()
elif params_loaded['type'] == config.TYPE_MNIST:
helper = ImageHelper(current_time=current_time,
params=params_loaded,
name=params_loaded.get('name', 'mnist'))
helper.load_data()
elif params_loaded['type'] == config.TYPE_TINYIMAGENET:
helper = ImageHelper(current_time=current_time,
params=params_loaded,
name=params_loaded.get('name', 'tiny'))
helper.load_data()
else:
helper = None
parser.add_argument('--name', dest='name', required=True)
args = parser.parse_args()
d = datetime.now().strftime('%b.%d_%H.%M.%S')
writer = SummaryWriter(log_dir=f'runs/{args.name}')
writer.add_custom_scalars(layout)
with open(args.params) as f:
params = yaml.load(f)
if params.get('model', False) == 'word':
helper = TextHelper(current_time=d, params=params, name='text')
helper.corpus = torch.load(helper.params['corpus'])
logger.info(helper.corpus.train.shape)
else:
helper = ImageHelper(current_time=d, params=params, name='utk')
logger.addHandler(
logging.FileHandler(filename=f'{helper.folder_path}/log.txt'))
logger.addHandler(logging.StreamHandler())
logger.setLevel(logging.DEBUG)
logger.info(f'current path: {helper.folder_path}')
batch_size = int(helper.params['batch_size'])
num_microbatches = int(helper.params['num_microbatches'])
lr = float(helper.params['lr'])
momentum = float(helper.params['momentum'])
decay = float(helper.params['decay'])
epochs = int(helper.params['epochs'])
S = float(helper.params['S'])
z = float(helper.params['z'])
sigma = z * S
dp = helper.params['dp']
parser.add_argument('--name', dest='name', required=True)
args = parser.parse_args()
d = datetime.now().strftime('%b.%d_%H.%M.%S')
writer = SummaryWriter(log_dir=f'runs/{args.name}')
writer.add_custom_scalars(layout)
with open(args.params) as f:
params = yaml.load(f)
if params.get('model', False) == 'word':
helper = TextHelper(current_time=d, params=params, name='text')
helper.corpus = torch.load(helper.params['corpus'])
logger.info(helper.corpus.train.shape)
else:
helper = ImageHelper(current_time=d, params=params, name=args.name)
logger.addHandler(
logging.FileHandler(filename=f'{helper.folder_path}/log.txt'))
logger.addHandler(logging.StreamHandler())
logger.setLevel(logging.DEBUG)
logger.info(f'current path: {helper.folder_path}')
batch_size = int(helper.params['batch_size'])
num_microbatches = int(helper.params['num_microbatches'])
lr = float(helper.params['lr'])
momentum = float(helper.params['momentum'])
decay = float(helper.params['decay'])
epochs = int(helper.params['epochs'])
S = float(helper.params['S'])
z = float(helper.params['z'])
sigma = z * S
dp = helper.params['dp']
res_orb_dir = os.path.join(
PATH_TO_RESULT_DIR, filename + str(resolution),
"ORB")
res_orb_data_dir = os.path.join(
res_orb_dir, "data_" + str(j))
if not os.path.exists(res_orb_dir):
os.makedirs(res_orb_dir, exist_ok=True)
if not os.path.exists(res_orb_data_dir):
os.makedirs(res_orb_data_dir, exist_ok=True)
if resolution < 1:
# downsize resolution
print("downsizing resolution to " +
str(resolution * 100) + " percent")
image_helper = ImageHelper()
image_helper.resize_directory(
os.path.join(PATH_TO_TMP_DIR, filename, "mav0",
"cam0", "data"),
int(752 * resolution), int(480 * resolution))
cg = CommandGenerator()
command = cg.orb(filename=filename,
path_to_orb=PATH_TO_ORB_SLAM,
path_to_data=PATH_TO_TMP_DIR,
path_to_config=PATH_TO_CONFIG,
dataset=dataset,
resolution=resolution)
print("Running ORB slam on {}!".format(filename))
t1 = time.perf_counter()
process = subprocess.Popen(command, shell=True)
def __init__(self, img_shape, latent_dim, num_classes, optimizer=Adam(lr=1e-4), dataset='', img_helper=ImageHelper()):
self.img_helper = img_helper
self.img_shape = img_shape
self.latent_dim = latent_dim
self.num_classes = num_classes
self.optimizer = optimizer
self.dataset = dataset
parser.add_argument('-p', '--params', dest='params')
args = parser.parse_args()
with open(f'./{args.params}', 'r') as f:
params_loaded = yaml.load(f)
current_time = datetime.now().strftime('%b.%d_%H.%M.%S')
# Loan dataset
if params_loaded['type'] == config.TYPE_LOAN:
helper = LoanHelper(current_time=current_time, params=params_loaded,
name=params_loaded.get('name', 'loan'))
helper.load_data(params_loaded)
# Image Datasets
elif params_loaded['type'] in [config.TYPE_CIFAR, config.TYPE_MNIST, config.TYPE_TINYIMAGENET]:
helper = ImageHelper(current_time=current_time, params=params_loaded,
name=params_loaded.get('name'))
helper.load_data()
# # Loan dataset
# if params_loaded['type'] == config.TYPE_LOAN:
# helper = LoanHelper(current_time=current_time, params=params_loaded,
# name=params_loaded.get('name', 'loan'))
# helper.load_data(params_loaded)
# # CIFAR dataset
# elif params_loaded['type'] == config.TYPE_CIFAR:
# helper = ImageHelper(current_time=current_time, params=params_loaded,
# name=params_loaded.get('name', 'cifar'))
# helper.load_data()
# # MNIST dataset
ai_model = model(model_file_path, labels_file_path)
# Initialize video reader
video_file_path = '../Videos/01.mp4'
video_reader = videoReader(video_file_path)
# Detection and preview parameters
score_threshold = 0.4
delay_between_frames = 5
# Perform object detection in the video sequence
while (True):
# Get frame from the video file
frame = video_reader.read_next_frame()
# If frame is None, then break the loop
if (frame is None):
break
# Perform detection
results = ai_model.detect_people(frame, score_threshold)
# Get centers of the bounding boxes (rectangle centers)
rectangle_centers = analyzer.get_rectangle_centers(results)
# Draw centers before displaying results
imgHelper.draw_rectangle_centers(frame, rectangle_centers)
# Display detection results
imgHelper.display_image_with_detected_objects(frame, results,
delay_between_frames)
def __init__(self, camera_overlay, resolution):
self.camera_overlay = camera_overlay
self.resolution = resolution
self.image_helper = ImageHelper(resolution)
self.main_screen_img = None
# Add reference to Part_03 (assuming the code is executed from Part_04 folder)
import sys
sys.path.insert(1, '../Part_03/')
from inference import Inference as model
from image_helper import ImageHelper as imgHelper
from camera import Camera as camera
if __name__ == "__main__":
# Load and prepare model
model_file_path = '../Models/01_model.tflite'
labels_file_path = '../Models/02_labels.txt'
# Initialize model
ai_model = model(model_file_path, labels_file_path)
# Initialize camera
camera_capture = camera()
# Capture frame and perform inference
camera_frame = camera_capture.capture_frame(False)
score_threshold = 0.5
results = ai_model.detect_objects(camera_frame, score_threshold)
# Display results
imgHelper.display_image_with_detected_objects(camera_frame, results)
import numpy as np from keras.datasets import fashion_mnist from image_helper import ImageHelper from gan import GAN (X, _), (_, _) = fashion_mnist.load_data() X_train = X / 127.5 - 1. X_train = np.expand_dims(X_train, axis=3) image_helper = ImageHelper() generative_advarsial_network = GAN(X_train[0].shape, 100, image_helper) generative_advarsial_network.train(30000, X_train, batch_size=32)
class PhotoboothController:
def __init__(self, camera, output_path, app):
self.output_path = output_path
self.last_file_path = None
self.app = app
self.camera = camera
self.camera_overlay = CameraOverlay(self.camera)
self.printer = Printer()
self.green_btn = Button(GREEN_BTN_PIN)
self.red_btn = Button(RED_BTN_PIN)
self.green_btn.when_pressed = self.pressed_capture_button
self.red_btn.when_pressed = self.pressed_reject_print_button
self.ui = UI(self.camera_overlay, camera.picam.resolution)
self.image_helper = ImageHelper(camera.picam.resolution)
self.rotary = Rotary(ROTARY_CLK_PIN, ROTARY_DT_PIN, upperBound=5)
self.camera.show_preview(True)
self.ui.show_main_screen()
self.waiting_for_confirm = False
self.busy = False
def _reset_state(self):
self.waiting_for_confirm = False
self.busy = False
self.rotary.clearCallback()
self.rotary.resetCount()
def pressed_capture_button(self):
print("\ncapture")
output_path = None
if self.waiting_for_confirm:
self.printer.printFile(self.last_file_path,
copies=self.rotary.getValue())
self.ui.clear_screen()
self.ui.show_main_screen()
self._reset_state()
elif not self.busy:
self.busy = True
# show countdowns
self.ui.clear_screen()
self.ui.show_countdown()
# flash lights
# self._flash()
# take photo
self.last_file_path = self._capture()
# confirm with user
self._confirm_print(self.last_file_path)
self.waiting_for_confirm = True
def pressed_reject_print_button(self):
if self.waiting_for_confirm:
print("\nreject")
self.last_file_path = None
self.ui.clear_screen()
self.ui.show_main_screen()
self._reset_state()
def _confirm_print(self, path):
preview_image = self.image_helper.load_image(path)
self.camera_overlay.add_overlay(preview_image)
self.ui.show_confirm_screen(1)
self.rotary.registerCallback(self.ui.update_confirm_screen)
def _flash(self):
pass
#img = self.image_helper.create_flash_image()
#self.camera_overlay.add_overlay(img)
#time.sleep(0.2)register
#self.camera_overlay.remove_top_overlay()
def _capture(self):
output_path = self._new_output_path()
print("saving photo to " + output_path)
self.camera.capture(output_path)
return output_path
def _new_output_path(self):
output_path = self.output_path + "image-%d-%mT%H:%M.png"
return strftime(output_path, gmtime())
driver.switch_to.default_content()
driver.switch_to.frame(driver.find_element_by_xpath("//iframe[contains(@src,'enquete')]"))
driver.switch_to.frame(driver.find_element_by_xpath("//iframe[contains(@title,'challenge')]"))
if check_exists_by_name(driver, "rc-imageselect-desc-no-canonical"):
text_tag = driver.find_element_by_class_name("rc-imageselect-desc-no-canonical")
sentence_tag = text_tag.find_element_by_tag_name("strong")
sentence_text = sentence_tag.text
sentence = sentence_text.split()
image_tag = driver.find_element_by_xpath("//img[contains(@class,'rc-image')]")
image_url = image_tag.get_attribute('src')
image_aux = ImageHelper(image_url)
image_aux.generate_images()
api = ClarifaiApi(app_id=clarifai_client_settings['id'],
app_secret=clarifai_client_settings['secret'],
language=clarifai_client_language)
tiles = driver.find_element_by_id("rc-imageselect-target")
table = tiles.find_element_by_tag_name("table")
tds = table.find_elements_by_tag_name("td")
if len(tds) == 9: # prevent error when recaptcha put more than 9 pics
for idx, td in enumerate(tds):
response = api.tag_images(image_aux.get_image(idx+1))
results = response['results'][0]['result']['tag']['classes']
def train(self, sess):
# loss summaries
d_summary_op = tf.summary.merge([
tf.summary.histogram("d_real_prob", tf.sigmoid(self.real_logits)),
tf.summary.histogram("d_fake_prob", tf.sigmoid(self.fake_logits)),
tf.summary.scalar("d_loss_fake", self.d_loss_fake),
tf.summary.scalar("d_loss_real", self.d_loss_real),
tf.summary.scalar("d_loss", self.d_loss)
],
name="discriminator_summary")
g_summary_op = tf.summary.merge([
tf.summary.histogram("g_prob", tf.sigmoid(self.fake_logits)),
tf.summary.scalar("g_loss", self.g_loss),
tf.summary.image("gen_images", self.fake_imgs)
],
name="generator_summary")
self.summary_dir = os.path.abspath(
os.path.join(self.hparams.checkpoint_dir, "summary"))
summary_writer = tf.summary.FileWriter(self.summary_dir, sess.graph)
image_helper = ImageHelper()
sess.run(tf.global_variables_initializer())
for num_epoch, num_batch, batch_images in image_helper.iter_images(
dirname=self.hparams.data_dir,
batch_size=self.batch_size,
epoches=self.epoches):
if (num_epoch == 0) and (num_batch < self.hparams.d_pretrain):
# pre-train discriminator
_, current_step, d_loss, d_accuarcy = sess.run(
[
self.d_optim, self.global_step, self.d_loss,
self.d_accuarcy
],
feed_dict={
self.rand_noises:
np.random.normal(
size=[self.batch_size, self.noise_dim]),
self.real_imgs:
batch_images
})
if current_step == self.hparams.d_pretrain:
tf.logging.info("==== pre-train ==== current_step:{}, d_loss:{}, d_accuarcy:{}"\
.format(current_step, d_loss, d_accuarcy))
else:
# optimize discriminator
_, current_step, d_loss, d_accuarcy = sess.run(
[
self.d_optim, self.global_step, self.d_loss,
self.d_accuarcy
],
feed_dict={
self.rand_noises:
np.random.normal(
size=[self.batch_size, self.noise_dim]),
self.real_imgs:
batch_images
})
# optimize generator
if current_step % self.hparams.d_schedule == 0:
_, g_loss = sess.run(
[self.g_optim, self.g_loss],
feed_dict={
self.rand_noises:
np.random.normal(
size=[self.batch_size, self.noise_dim])
})
# summary
if current_step % self.hparams.log_interval == 0:
d_summary_str, g_summary_str = sess.run(
[d_summary_op, g_summary_op],
feed_dict={
self.rand_noises:
np.random.normal(
size=[self.batch_size, self.noise_dim]),
self.real_imgs:
batch_images
})
summary_writer.add_summary(d_summary_str, current_step)
summary_writer.add_summary(g_summary_str, current_step)
tf.logging.info("step:{}, d_loss:{}, d_accuarcy:{}, g_loss:{}"\
.format(current_step, d_loss, d_accuarcy, g_loss))
if (num_epoch > 0) and (num_batch == 0):
# generate images per epoch
tf.logging.info(
"epoch:{} === generate images and save checkpoint".format(
num_epoch))
fake_imgs = sess.run(
self.fake_imgs,
feed_dict={
self.rand_noises:
np.random.normal(
size=[self.batch_size, self.noise_dim])
})
image_helper.save_imgs(fake_imgs,
img_name="{}/fake-{}".format(
self.hparams.sample_dir, num_epoch))
# save model per epoch
self.saver.save(sess,
self.checkpoint_prefix,
global_step=num_epoch)
import common
from image_helper import ImageHelper as imgHelper
from inference import Inference as model
if __name__ == "__main__":
# Load and prepare model
model_file_path = '../Models/01_model.tflite'
labels_file_path = '../Models/02_labels.txt'
# Initialize model
ai_model = model(model_file_path, labels_file_path)
# Get input image
image = imgHelper.load_image('../Images/Lena.png')
# Detect objects
score_threshold = 0.5
results = ai_model.detect_objects(image, score_threshold)
# Display results
imgHelper.display_image_with_detected_objects(image, results)
class UI:
def __init__(self, camera_overlay, resolution):
self.camera_overlay = camera_overlay
self.resolution = resolution
self.image_helper = ImageHelper(resolution)
self.main_screen_img = None
def show_main_screen(self):
if self.main_screen_img == None:
self.main_screen_img = Image.new('RGBA', self.resolution,
(0, 0, 0, 0))
self.image_helper.create_circle_image(self.main_screen_img,
UIConst.BUTTON_DIA,
"green",
ImagePosition.LOWERLEFT,
yPadding=UIConst.YPADDING)
text = self.image_helper.create_button_text_image(
self.main_screen_img,
"Go!",
UIConst.FONT_SIZE,
UIConst.BUTTON_DIA,
UIConst.BUTTON_DIA,
ImagePosition.LOWERLEFT,
yPadding=UIConst.YPADDING)
self.camera_overlay.add_overlay(self.main_screen_img)
def show_confirm_screen(self, print_count):
img = Image.new('RGBA', self.resolution, (0, 0, 0, 0))
printLabel = "Copies: " + str(print_count)
self.image_helper.create_circle_image(img,
UIConst.BUTTON_DIA,
"red",
ImagePosition.LOWERLEFT,
yPadding=UIConst.YPADDING)
self.image_helper.create_circle_image(img,
UIConst.BUTTON_DIA,
"green",
ImagePosition.LOWERRIGHT,
yPadding=UIConst.YPADDING)
self.image_helper.create_button_text_image(img,
"Retake",
UIConst.FONT_SIZE,
UIConst.BUTTON_DIA,
UIConst.BUTTON_DIA,
ImagePosition.LOWERLEFT,
yPadding=UIConst.YPADDING)
self.image_helper.create_button_text_image(img,
"Print!",
UIConst.FONT_SIZE,
UIConst.BUTTON_DIA,
UIConst.BUTTON_DIA,
ImagePosition.LOWERRIGHT,
yPadding=UIConst.YPADDING)
self.image_helper.create_text_image(img,
printLabel,
UIConst.FONT_SIZE,
ImagePosition.UPPERCENTER,
yPadding=UIConst.YPADDING)
self.camera_overlay.add_overlay(img)
def update_confirm_screen(self, print_count):
print("update confirm screen with count: " + str(print_count))
# pretty hacky since we're tearing down the whole ui just to update a label
self.camera_overlay.remove_top_overlay()
self.show_confirm_screen(print_count)
def clear_screen(self):
self.camera_overlay.remove_overlays()
def show_countdown(self):
for i in range(3, 0, -1):
img = Image.new('RGBA', self.resolution, (0, 0, 0, 0))
self.image_helper.create_text_image(img, str(i), UIConst.FONT_SIZE,
ImagePosition.UPPERCENTER)
self.camera_overlay.add_overlay(img)
time.sleep(1)
self.camera_overlay.remove_top_overlay()
time.sleep(0.05)
return total_l, acc
if __name__ == '__main__':
print('Start training')
time_start_load_everything = time.time()
parser = argparse.ArgumentParser(description='PPDL')
parser.add_argument('--params', dest='params', default='params_words.json')
args = parser.parse_args()
with open(f'./{args.params}', 'r') as f:
params_loaded = yaml.load(f)
current_time = datetime.datetime.now().strftime('%b.%d_%H.%M.%S')
if params_loaded['type'] == "image":
helper = ImageHelper(current_time=current_time, params=params_loaded,
name=params_loaded.get('name', 'image'))
else:
helper = TextHelper(current_time=current_time, params=params_loaded,
name=params_loaded.get('name', 'text'))
helper.load_data()
helper.create_model()
### Create models
if helper.params['is_poison']:
helper.params['adversary_list'] = [0]+ \
random.sample(range(helper.params['number_of_total_participants']),
helper.params['number_of_adversaries']-1)
logger.info(f"Poisoned following participants: {len(helper.params['adversary_list'])}")
else:
helper.params['adversary_list'] = list()
# Initialize model
ai_model = model(model_file_path, labels_file_path)
# Initialize video reader
video_file_path = '../Videos/01.mp4'
video_reader = videoReader(video_file_path)
# Detection and preview parameters
score_threshold = 0.4
detect_only_people = True
delay_between_frames = 5
# Perform object detection in the video sequence
while (True):
# Get frame from the video file
frame = video_reader.read_next_frame()
# If frame is None, then break the loop
if (frame is None):
break
# Perform detection
if (detect_only_people):
results = ai_model.detect_people(frame, score_threshold)
else:
results = ai_model.detect_objects(frame, score_threshold)
# Display results
imgHelper.display_image_with_detected_objects(frame, results,
delay_between_frames)