def get_test_images(test_file, test_folder):
print("Getting test images...")
if not isdir(test_folder):
os.mkdir(test_folder)
pic_to_faces = OrderedDict()
# Read in test_file contents
with open(test_file, "r") as csvfile:
test_items = DictReader(
csvfile, fieldnames=['url', 'file_name', 'num_faces', 'md5_hash'])
for row in test_items:
test_file_name = join(test_folder, row['file_name'])
try:
# File has already been downloaded
if isfile(test_file_name) and row['md5_hash'] == file_digest(
test_file_name):
pic_to_faces[test_file_name] = int(row['num_faces'])
#print("Verified {0}".format(test_file_name))
continue
# Download file and test hash
else:
if py3:
urlretrieve(row['url'], test_file_name)
else:
response = urlopen(row['url'])
with open(test_file_name, "wb") as local_file:
local_file.write(response.read())
if row['md5_hash'] != file_digest(test_file_name):
os.remove(test_file_name)
print(
"Warning - {0} did not match expected hash".format(
test_file_name))
else:
pic_to_faces[test_file_name] = int(row['num_faces'])
#print("Downloaded {0}".format(test_file_name))
except:
print("Error - ", exc_info())
continue
print("{0} test images verified\n".format(len(pic_to_faces)))
return pic_to_faces
def process_image(image_file,
margin,
min_size,
threshold,
scale_factor,
reduceby,
pnet,
rnet,
onet,
verbose=False):
filename = image_file.split('/')[-1]
file_content_hash = file_digest(image_file)
image = cv2.imread(image_file)
frame_number = -1
# Find bounding boxes for face chips in this image
face_locations, num_detections = identify_chips(image, frame_number,
margin, min_size,
threshold, scale_factor,
reduceby, pnet, rnet, onet)
# Only save pickle if faces were detected
if num_detections > 0:
results = (filename, file_content_hash, [face_locations])
write_out_pickle(filename, results, "/bboxes", "mtcnn", "bboxes")
if verbose:
print("{0} face detections in {1}".format(num_detections, filename))
def process_image(image_file,
reduceby,
prob_thresh,
nms_thresh,
use_gpu,
verbose=False):
filename = image_file.split('/')[-1]
file_content_hash = file_digest(image_file)
image = cv2.imread(image_file)
frame_number = -1
# Find bounding boxes for face chips in this image
face_locations, num_detections = identify_chips(image, frame_number,
reduceby, prob_thresh,
nms_thresh, use_gpu)
# Only save pickle if faces were detected
if num_detections > 0:
results = (filename, file_content_hash, [face_locations])
write_out_pickle(filename, results, "/bboxes", "tinyface", "bboxes")
if verbose:
print("{0} face detections in {1}".format(num_detections, filename))
def process_video(image_file, threshold, nms, reduceby, net, every):
frame_number = 0
num_detections = 0
filename = image_file.split('/')[-1]
#camera = cv2.VideoCapture(image_file)
#capture_length = int(camera.get(cv2.CAP_PROP_FRAME_COUNT))
capture_length = get_movie_length(image_file)
progress = tqdm(total=capture_length)
file_content_hash = file_digest(image_file)
combined_face_locations = list()
keep_going = True
first = True
while keep_going:
if not first:
if (every + frame_number) > capture_length:
keep_going = False
progress.close()
break
frame_number += every
#camera.set(1, frame_number)
progress.update(every)
else:
first = False
#keep_going, image = camera.read()
keep_going, image = get_frame_inefficient(image_file, frame_number)
# only face detect every once in a while
progress.set_description(
'Processing video: {0} detections: {1}'.format(
filename[0:30] + "...", num_detections))
progress.refresh()
# verify that there is a video frame to process
if image is None:
progress.refresh()
progress.write('end of capture:IMG')
progress.close()
break
if frame_number > capture_length:
progress.refresh()
progress.write('end of capture:Length')
progress.close()
break
if not keep_going:
progress.refresh()
progress.write('end of capture:camera.read')
progress.close()
break
# Find bounding boxes for face chips in this frame
face_locations, detections = identify_chips(image, frame_number,
threshold, nms, reduceby,
net)
if detections > 0:
combined_face_locations += [face_locations]
num_detections += detections
# Only save pickle if faces were detected
if num_detections > 0:
results = (filename, file_content_hash, combined_face_locations)
write_out_pickle(filename, results, "/bboxes", "frcnn", "bboxes")
def main(jitters=1, tolerance=0.6, chip_size=160, verbose=False):
# Gather file names
files = [item for item in glob.glob(
'/bboxes/*') if not os.path.isdir(item)]
if verbose:
durations = list()
kickoff = time()
for f in files:
print('Processing:', f.split('/')[-1])
filename, file_content_hash, bounding_boxes = pickle.load(
open(f, "rb"))
file_with_path = join('/media', filename)
# Verify original file exists on disk and has same content hash
if os.path.isfile(file_with_path) and file_digest(
file_with_path) == file_content_hash:
if len(bounding_boxes) > 0:
# only allow verbose flag for images
if verbose and bounding_boxes[0][0] == -1:
start = time()
extract_chips(
filename,
bounding_boxes,
chip_size,
file_content_hash,
tolerance,
jitters,
verbose)
duration = time() - start
durations.append(duration)
print("{0} seconds to process {1}\n".format(
'%.3f' % duration, f.split('/')[-1]))
else:
extract_chips(
filename,
bounding_boxes,
chip_size,
file_content_hash,
tolerance,
jitters)
else:
print("There were no faces detected in {0}".format(f))
else:
print(
"\tWarning - {0} has bounding box file in /bboxes but was not found in /media.\n".format(filename))
sys.stdout.flush()
sys.stderr.flush()
if verbose and len(durations) > 0:
average = sum(durations) / len(durations)
print(
"\nAverage elapsed time to vectorize faces in an image = {0}".format(
'%.3f' %
average))
print(
"Total time to vectorize faces in {0} images = {1}".format(
len(durations), '%.3f' %
(time() - kickoff)))
def get_iou_images(iou_url, test_folder, num_test_images=20, seed=41):
print("Downloading list of test images...")
if not isdir(test_folder):
os.mkdir(test_folder)
pic_to_bbox = OrderedDict()
# Go get the iou_url file
iou_file_name = "../test_data/iou_test_file.txt"
try:
if py3:
urlretrieve(iou_url, iou_file_name)
else:
response = urlopen(iou_url)
with open(iou_file_name, "wb") as local_file:
local_file.write(response.read())
except:
print("Error retrieving {0} - {1}".format(iou_url, exc_info()))
quit()
# Read in iou_file_name contents
with open(iou_file_name, "r") as csvfile:
iou_items = DictReader(
csvfile,
fieldnames=['person', 'imagenum', 'url', 'rect', 'md5_hash'],
delimiter='\t')
marker = 0
candidate_list = list()
for item in iou_items:
# Skip first two lines of header info
if item['md5_hash'] is None:
continue
candidate_list.append(item)
# Sample for variety of faces
random.seed(seed)
candidate_list = random.sample(candidate_list, num_test_images * 20)
print("Looking for previously downloaded test images...")
pop_list = list()
for i in range(len(candidate_list)):
test_file_name = join(
test_folder, candidate_list[i]['person'].replace(" ", "") +
candidate_list[i]['imagenum'] + ".iou")
if isfile(test_file_name) and candidate_list[i][
'md5_hash'] == file_digest(test_file_name):
bbox = [
int(coordinate)
for coordinate in candidate_list[i]['rect'].split(",")
]
pic_to_bbox[test_file_name] = bbox
pop_list.append(i)
# Remove already downloaded files from candidate_list
for index_to_pop in pop_list:
candidate_list.pop(index_to_pop)
print("Downloading test images (slow)...")
for row in candidate_list:
test_file_name = join(
test_folder,
row['person'].replace(" ", "") + row['imagenum'] + ".iou")
try:
# File has already been downloaded
#if isfile(test_file_name) and row['md5_hash'] == file_digest(test_file_name):
# bbox = [int(coordinate) for coordinate in row['rect'].split(",")]
# pic_to_bbox[test_file_name] = bbox
#print("Verified {0}".format(test_file_name))
# continue
# Download file and test hash
#else:
if py3:
urlretrieve(row['url'], test_file_name)
else:
response = urlopen(row['url'])
with open(test_file_name, "wb") as local_file:
local_file.write(response.read())
if row['md5_hash'] != file_digest(test_file_name):
os.remove(test_file_name)
#print("Warning - {0} did not match expected hash".format(test_file_name))
else:
bbox = [
int(coordinate)
for coordinate in row['rect'].split(",")
]
pic_to_bbox[test_file_name] = bbox
#print("Downloaded {0}".format(test_file_name))
except:
#print("Error - ", exc_info())
continue
# Check is enough test images downloaded
if len(pic_to_bbox) % 5 == 0 and len(pic_to_bbox) > marker:
print("{0} of {1} test images downloaded and verified".format(
len(pic_to_bbox), num_test_images))
marker = len(pic_to_bbox)
if len(pic_to_bbox) == num_test_images:
break
#print("{0} test images verified\n".format(len(pic_to_bbox)))
return pic_to_bbox
def main(model,
dlibFacePredictor,
use_gpu=False,
tolerance=0.6,
chip_size=96,
verbose=False):
print("Loading Facenet Model...")
sys.stdout.flush()
sys.stderr.flush()
# Load the Facenet model
if use_gpu:
net = TorchNeuralNet(model, chip_size, cuda=True)
else:
net = TorchNeuralNet(model, chip_size, cuda=False)
# Load dlib image aligner
aligner = AlignDlib(dlibFacePredictor)
# Gather file names
files = [
item for item in glob.glob('/bboxes/*') if not os.path.isdir(item)
]
if verbose:
durations = list()
kickoff = time()
for f in files:
print('Processing:', f.split('/')[-1])
filename, file_content_hash, bounding_boxes = pickle.load(open(
f, "rb"))
file_with_path = join('/media', filename)
# Verify original file exists on disk and has same content hash
if os.path.isfile(file_with_path) and file_digest(
file_with_path) == file_content_hash:
if len(bounding_boxes) > 0:
# only allow verbose on images
if verbose and bounding_boxes[0][0] == -1:
start = time()
extract_chips(filename, bounding_boxes, chip_size,
file_content_hash, tolerance, net, aligner,
verbose)
duration = time() - start
durations.append(duration)
print("{0} seconds to process {1}\n".format(
'%.3f' % duration,
f.split('/')[-1]))
else:
extract_chips(filename, bounding_boxes, chip_size,
file_content_hash, tolerance, net, aligner)
else:
print("There were no faces detected in {0}".format(f))
else:
print(
"\tWarning - {0} has bounding box file in /bboxes but was not found in /media.\n"
.format(filename))
sys.stdout.flush()
sys.stderr.flush()
if verbose and len(durations) > 0:
average = sum(durations) / len(durations)
print("\nAverage elapsed time to vectorize faces in an image = {0}".
format('%.3f' % average))
print("Total time to vectorize faces in {0} images = {1}".format(
len(durations), '%.3f' % (time() - kickoff)))
def main(model,
use_gpu=False,
gpu_memory_fraction=0.8,
tolerance=0.6,
chip_size=160,
verbose=False):
if use_gpu:
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
else:
sess = tf.Session()
# Load the model
print("Loading Facenet Model...")
sys.stdout.flush()
sys.stderr.flush()
facenet.load_model(model, sess)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
phase_train_placeholder = tf.get_default_graph().get_tensor_by_name(
"phase_train:0")
# Gather files names
files = [
item for item in glob.glob('/bboxes/*') if not os.path.isdir(item)
]
if verbose:
durations = list()
kickoff = time()
for f in files:
print('Processing:', f.split('/')[-1])
filename, file_content_hash, bounding_boxes = pickle.load(open(
f, "rb"))
file_with_path = join('/media', filename)
# Verify original file exists on disk and has same content hash
if os.path.isfile(file_with_path) and file_digest(
file_with_path) == file_content_hash:
if len(bounding_boxes) > 0:
if verbose and bounding_boxes[0][0] == -1:
start = time()
extract_chips(filename, bounding_boxes, chip_size,
file_content_hash, tolerance,
images_placeholder, phase_train_placeholder,
embeddings, sess, verbose)
duration = time() - start
durations.append(duration)
print("{0} seconds to process {1}\n".format(
'%.3f' % duration,
f.split('/')[-1]))
else:
extract_chips(filename, bounding_boxes, chip_size,
file_content_hash, tolerance,
images_placeholder, phase_train_placeholder,
embeddings, sess)
else:
print("There were no faces detected in {0}".format(f))
else:
print(
"\tWarning - {0} has bounding box file in /bboxes but was not found in /media.\n"
.format(filename))
sys.stdout.flush()
sys.stderr.flush()
sess.close()
if verbose and len(durations) > 0:
average = sum(durations) / len(durations)
print("\nAverage elapsed time to vectorize faces in an image = {0}".
format('%.3f' % average))
print("Total time to vectorize faces in {0} images = {1}".format(
len(durations), '%.3f' % (time() - kickoff)))