def test(label_dict, image_ids, args):
"""
Test the retrieval system. For each input in image_ids, generate a list of IDs returned for
that image.
:param label_dict: Dictionary containing labels (and their confidence) for each image
:param image_ids: List of image IDs. Each element is assumed to be an entry in the test set,
and is located at "pics/<image id>.jpg" in the test path
:param args: Run-time arguments
:return: A dictionary with keys equal to the images in the queries list,
and values a list of image IDs retrieved for that input
"""
log_header('Starting image retrieval preparations')
queries = []
if args.test_image is not None:
queries.append(args.test_image)
else:
logging.info('Generating random test queries')
number_of_images = len(image_ids)
# Generate random queries to use in the test procedure
for i in range(args.k):
queries.append(image_ids[random.randint(0, number_of_images - 1)])
# Calculate score for the retrieved images
calculate_score(label_dict, queries,
retrieve_similar_images(queries, args))
def main(options, logger=None):
if logger == None:
logger = Logger("download.log", "download.py", True, True)
else:
logger.context = "download.py"
try:
query = utils.getQuery()
logger.context = "download.py"
if __name__ == "__main__":
utils.log_header(logger, DESCRIPTION,
options.get("verbose", False))
logger.debug("Options:", options.get("verbose", False))
for i in options:
logger.debug(i + ": " + str(options[i]),
options.get("verbose", False))
download_query(query, options, logger)
# move all downloaded videos in videos directory
repoDir = utils.repoPath()
fs = os.listdir(repoDir)
for f in fs:
if utils.is_video(f):
os.rename(f, "videos/" + f)
logger.success("All files downloaded.",
not options.get("silent", False))
except Exception as e:
logger.handle_exception(e)
raise e
def main(options, logger=Logger()):
logger.context = "transfer.py"
VERBOSE = options.get("verbose", False)
if __name__ == "__main__":
utils.log_header(logger, utils.DESCRIPTION, VERBOSE)
for i in utils.getVideos():
videoPath = utils.videosPath()
if not (os.listdir(str(Path.home()) + "/" +
videoPath).__contains__(i)):
logger.plain("copying to (home)/" + videoPath + ": " + str(i),
VERBOSE)
shutil.copyfile("videos/" + i,
str(Path.home()) + "/" + videoPath + "/" + i)
for i in utils.getMP3s():
musicPath = utils.musicPath()
if not (os.listdir(str(Path.home()) + "/" +
musicPath).__contains__(i)):
logger.plain("copying to (home)/" + musicPath + ": " + str(i),
VERBOSE)
shutil.copyfile("music/mp3/" + i,
str(Path.home()) + "/" + musicPath + "/" + i)
logger.success("Copied all pending files.",
not options.get("silent", False))
logger.save()
def calculate_score(label_dict, queries, results):
"""
Calculate score for queried images
:param label_dict: Dictionary of labels, keys are image IDs
:param queries: List of image ids to query
:param results: Retrieved image ids for each image in queries
:return: Total score
"""
total_score = 0.0
log_header('Individual image scores')
# Calculate score for all images
for image_id in queries:
if image_id in results.keys():
# Run the score function
image_score = score(label_dict=label_dict,
target=image_id,
selection=results[image_id])
else:
logging.error('No result generated for %s' % image_id)
image_score = 0.0
total_score += image_score
logging.info('%s: %8.6f' % (image_id, image_score))
log_header('Average score over %d images: %10.8f' %
(len(queries), total_score / len(queries)))
return total_score
def train(args):
"""
Run the training procedure
:param args: Run-time arguments
"""
log_header('Training network')
train_retriever(args)
def train_feature_model(label_dict, args):
"""
Train the feature extractor
:param label_dict: Dictionary containing labels (and their confidence)
:param args: Run-time arguments
"""
log_header('Training feature model')
run_network(feature_network(args),
args.feature_model,
args,
training_data=(generate_training_batch, {
'label_dict': label_dict,
'args': args
}))
def main(options, logger=Logger()):
VERBOSE = options.get("verbose", False)
logger.context = "convert.py"
if __name__ == "__main__":
utils.log_header(logger, DESCRIPTION, VERBOSE)
logger.debug("Bitrate: " + str(options.get("bitrate", "(error)")), VERBOSE)
utils.convertFiles(options.get("bitrate", 320),
options.get("silent", False))
logger.plain(
"Converted files. Now extracting thumbnails from videos...", VERBOSE)
utils.extractThumbnails(options.get("silent", False))
logger.plain(
"Extracted thumbnails from videos, now writing them to mp3-files...", VERBOSE)
utils.writeThumbnails()
logger.success("All files converted.", not options.get("silent", False))
logger.save()
def generate_features(args):
"""
Generate and save feature values for all images in training path
:param args: Run-time arguments
"""
log_header('Generating features')
images = []
# Add all images in training path
for image_path, _, image_id in get_images_in_path(args.train_path):
images.append((image_path, image_id))
# Generate features for all images
run_network(feature_network(args),
args.feature_model,
args,
train=False,
generating_data=images,
save_path=args.features)
def run_network(network,
model_name,
args,
train=True,
training_data=None,
value=None,
generating_data=None,
save_path=None):
"""
Run a neural network. Can either train weights, evaluate input values, or generate new features
for images.
When generating features, the features are saved in batches to pickle files
:param network: Network to run
:param model_name: Name of the checkpoint file. Either for saving or loading weights
:param args: Run-time arguments
:param train: Whether to train or evaluate
:param training_data: Data used for training
:param value: Value used for evaluating
:param generating_data: Data used for generating new features
:param save_path: Path to save new features
:return: Evaluated value
"""
# Set up session
with tf.Session() as sess:
# Extract variables from network
x, y, output_layer, cost_function, optimizer = network
# Initial variables and set up saver
sess.run(tf.initialize_all_variables())
saver = tf.train.Saver()
# Enable training if wanted
if train:
logging.info('Training model')
try:
# Iterate training epochs
for epoch in range(args.training_epochs):
log_header('Epoch: %d' % epoch)
# Iterate batches in epoch
for batch in range(0, args.number_of_batches):
logging.info('Epoch: %d. Batch: %d' % (epoch, batch))
# Generate training batch
x_, y_ = training_data[0](**training_data[1])
# Run the optimizer
sess.run([optimizer, cost_function],
feed_dict={
x: x_,
y: y_
})
logging.info('Training complete')
except KeyboardInterrupt:
logging.error('Training aborted')
finally:
logging.info('Saving model')
# Save trained weights
saver.save(sess, model_name)
logging.debug('Model saved to %s' % model_name)
else:
# Import and restore trained weights
saver = tf.train.import_meta_graph('%s.meta' % model_name)
saver.restore(sess, model_name)
# Evaluate value if supplied
if value is not None:
res = []
# Evaluate in batches in order not to deplete memory
for i in range(0, len(value), args.batch_size):
output = np.squeeze(
sess.run([output_layer],
feed_dict={x: value[i:i + args.batch_size]}))
if len(output) != args.batch_size:
output = [output]
res.extend(output)
return res
# Set up batches for saving features
save_batch_number_of_batches = 50
save_batch_size = args.batch_size * save_batch_number_of_batches
total_testing_data = len(generating_data)
total_save_batches = math.ceil(total_testing_data /
save_batch_size)
# Iterate every feature saving batch
for save_batch_number in range(total_save_batches):
logging.error('Batch %d of %d' %
(save_batch_number + 1, total_save_batches))
# Set up batch variables
save_batch_offset = save_batch_number * save_batch_size
save_batch_name = '%s.%d' % (args.features, save_batch_number)
save_batch_features = {}
# Iterate smaller batches
for batch_number in range(save_batch_number_of_batches):
batch_offset = save_batch_offset + batch_number * save_batch_number_of_batches
if batch_offset >= total_testing_data:
break
# Extract training batch
batch = generating_data[batch_offset:batch_offset +
save_batch_number_of_batches]
inputs = []
# Pre-process images in batch
for image_path, image_id in batch:
inputs.append(
preprocess_image(
'%s/%s.jpg' % (image_path, image_id), args))
# Generate features
features = np.squeeze(
sess.run([output_layer], feed_dict={x: inputs}))
# Add features to batch
for i in range(len(batch)):
_, image_id = batch[i]
save_batch_features[image_id] = features[i]
# Save batch to file
with open(save_batch_name, 'wb') as f:
pickle.dump(save_batch_features, f)
# Start THREAD
#
poll_thread = THREAD(joy_range, logfilename, joy_input)
poll_thread.daemon = True
poll_thread.start()
##############################
# Start DISPLAY
mode = DISPLAY('USB connection', poll_thread, task_info, FPS, fullscreen_on, debug_mode)
mode.task_info['subject_id'] = subject_id
#
# write header
#
log_header(poll_thread.logfile, poll_thread.joy_status, mode.task_info, poll_thread.joy_range, min_median_ref)
#
# Start Labjack
#
if param['start_emg'] or param['run_stimulation'] or param['monitor_emg']:
import labjack_interface as lji
lji.reset()
mode.task_info['description'] = 'start labjack'
mode.task_info['time'] = datetime.now().strftime("%H:%M:%S.%f")
log(poll_thread.logfile, poll_thread. joy_status,mode.task_info)
target_on = False
#
# CREATION TARGETS:
#
poll_thread = THREAD(joy_range, logfilename, joy_input)
poll_thread.daemon = True
poll_thread.start()
##############################
# Start DISPLAY
mode = DISPLAY('USB connection', poll_thread, task_info, FPS, fullscreen_on,
debug_mode)
mode.task_info['subject_id'] = subject_id
#
# write header
#
log_header(poll_thread.logfile, poll_thread.joy_status, mode.task_info,
poll_thread.joy_range, min_median_ref)
#
# Start Labjack
#
if param['start_emg'] or param['run_stimulation'] or param['monitor_emg']:
import labjack_interface as lji
lji.reset()
mode.task_info['description'] = 'start labjack'
mode.task_info['time'] = datetime.now().strftime("%H:%M:%S.%f")
log(poll_thread.logfile, poll_thread.joy_status, mode.task_info)
target_on = False
#
# CREATION TARGETS:
#
# Start THREAD
#
poll_thread = THREAD(joy_range, logfilename, joy_input)
poll_thread.daemon = True
poll_thread.start()
#
# Start DISPLAY
#
mode = DISPLAY('USB connection', poll_thread, task_info, FPS, fullscreen_on)
#
# write header
#
log_header(poll_thread.logfile, poll_thread.joy_status, mode.task_info,
joy_range)
#
# Start Labjack
#
if param['start_emg'] or param['run_stimulation'] or param['monitor_emg']:
import labjack_interface as lji
lji.reset()
mode.task_info['description'] = 'start labjack'
mode.task_info['time'] = datetime.now().strftime("%H:%M:%S.%f")
log(poll_thread.logfile, poll_thread.joy_status, mode.task_info)
target_on = False
#
# 6 : PSEUDO RANDOM
#
# start THREAD
#
poll_thread = THREAD(joy_range, logfilename, joy_input)
poll_thread.daemon = True
poll_thread.start()
#
# start DISPLAY
#
mode = DISPLAY('USB connection', poll_thread, task_info, FPS, fullscreen_on)
#
# write header
#
log_header(poll_thread.logfile, poll_thread.joy_status, mode.task_info)
#
# display calibration
#
mode.display.fill(colors.BLACK)
mode.display_text('CALIBRATION', colors.RED, 100, datetime.now(), 5)
mode.display_text(calibration_string[0], colors.RED, 50, datetime.now(), 2)
running_calibration = True
next_order = True
calibration_string_order = 0
while running_calibration:
action="store_true",
help=
"when wished format is not availabe, it will be automatically converted by default. This keeps the old format, too."
)
parser.add_argument("--filename",
"--fname",
nargs="?",
type=str,
help="filename to save the file under")
arguments = parser.parse_args()
args = vars(arguments)
VERBOSE = args.get("verbose", False)
log = Logger("workflow.log", "workflow.py", True, True)
utils.log_header(log, DESCRIPTION, VERBOSE)
try:
log.debug("Debug mode activated.", VERBOSE)
log.debug("Args: ", VERBOSE)
for i in args:
log.debug(i + ": " + str(args[i]), VERBOSE)
if not args.get("test", False):
download.main(args, logger=log)
if args.get("facerec", None) != None and not args.get("test", False):
if "all" in args.get("facerec", []):
# pass all videos from utils.getVideos() to facerec
facerec.main({"files": utils.getVideos()})
else:
# pass all videos from args to facerec
facerec.main({"files": args.get("facerec", [])})