def show_all_faces(args, svm_clf, knn_clf):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
for i,n in enumerate(faces.dict_by_name):
args.face = n
show_faces_by_name(args, svm_clf, knn_clf, faces, img_labels)
def export_face_imgs_only(args):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
for im in faces.dict_by_files:
rel_path = os.path.relpath(im, args.imgs_root)
symlinkname = os.path.join(args.outdir, rel_path)
xmp_path = os.path.splitext(im)[0] + '.xmp'
xmp_path_link = os.path.splitext(symlinkname)[0] + '.xmp'
names = faces.get_names(faces.dict_by_files[im])
if names.count('unknown') + names.count('deleted') + names.count(
'detected') == len(names):
if os.path.exists(symlinkname):
os.remove(symlinkname)
if os.path.exists(xmp_path_link):
os.remove(xmp_path_link)
else:
dirname = os.path.dirname(symlinkname)
if not os.path.exists(dirname):
os.makedirs(dirname)
if not os.path.islink(symlinkname):
os.symlink(im, symlinkname)
if not os.path.islink(xmp_path_link):
os.symlink(xmp_path, xmp_path_link)
def detect_faces(args):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
# initialize counters
total_faces = len(utils.get_images_in_dir_rec(args.input))
detect_faces_in_folder(args, faces, img_labels,
utils.get_images_in_dir_rec(args.input),
total_faces)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--face', type=str, required=True,
help="Face to show ('all' shows all faces).")
parser.add_argument('--svm', type=str, required=True,
help="Path to svm model file (e.g. svm.clf).")
parser.add_argument('--knn', type=str, required=True,
help="Path to knn model file (e.g. knn.clf).")
# parser.add_argument('--db', type=str, required=True,
# help="Path to folder with predicted faces (.csv files).")
parser.add_argument('--imgs_root', type=str, required=True,
help="Root directory of your image library.")
parser.add_argument('--mask_folder', type=str, required=False, default='',
help="Mask folder for faces. Only faces of images within this folder will be shown.")
parser.add_argument('--min_size', type=str, required=False, default=0,
help="Defines the min. size of a face. A face will be accepted if it is larger than min_size * img_height (img_width resp.)")
args = parser.parse_args()
# if not os.path.isdir(args.db):
# print('args.db is not a valid directory')
if os.path.isfile(args.knn):
with open(args.knn, 'rb') as f:
knn_clf = pickle.load(f)
else:
print('args.knn ({}) is not a valid file'.format(args.knn))
exit()
if os.path.isfile(args.svm):
with open(args.svm, 'rb') as f:
svm_clf = pickle.load(f)
else:
print('args.svm ({}) is not a valid file'.format(args.svm))
exit()
if os.path.isdir(args.face):
print('Showing detections of folder {}'.format(args.face))
show_faces_in_folder(args, svm_clf, knn_clf)
else:
if args.face in ['predicted', 'unconfirmed']:
print('Showing unconfirmed faces')
show_unconfirmed_faces(args, svm_clf, knn_clf)
else:
print('Showing detections of class {}'.format(args.face))
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
# with open(os.path.join(args.imgs_root, 'faces_single_file.bin'), 'wb') as fid:
# pickle.dump(tmp_faces, fid)
faces = utils.FACES(tmp_faces)
show_faces_by_name(args, svm_clf, knn_clf, faces, img_labels)
# show_all_faces(args, svm_clf, knn_clf)
print('Done.')
def tag_images(args):
# images = utils.get_images_in_dir_rec(args.input)
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
images = list(faces.dict_by_files.keys())
random.shuffle(images)
logging.basicConfig(level=logging.INFO,
format='%(levelname)s: %(message)s')
rekognition_client = boto3.client('rekognition')
for img in images:
pkl_path = img + '.pkl'
if os.path.isfile(pkl_path):
with open(pkl_path, 'rb') as fid:
img_label = pickle.load(fid)
else:
img_label = utils.IMG_LABELS(utils.get_timestamp(img))
img_label.path = img
# if not hasattr(img_label, 'gcloud_labels'):
# img_label.gcloud_labels = []
# if not hasattr(img_label, 'gcloud_objects'):
# img_label.gcloud_objects = []
# if not hasattr(img_label, 'gcloud_landmarks'):
# img_label.gcloud_landmarks = []
# if not hasattr(img_label, 'gcloud_web'):
# img_label.gcloud_web = []
# if len(img_label.gcloud_labels) != 0 or \
# len(img_label.gcloud_objects) != 0 or \
# len(img_label.gcloud_web) != 0 or \
# len(img_label.gcloud_landmarks):
# print('{} already tagged'.format(img))
# continue
print('tagging {}'.format(img))
labels = detect_labels_local_file(img)
# img_label.gcloud_labels = detect_labels(image)
# img_label.gcloud_objects = detect_objects(image)
# img_label.gcloud_web = detect_web(image)
# img_label.gcloud_landmarks = detect_landmarks(image)
if 1:
ocv_img = cv2.imread(img)
cv2.imshow("image", ocv_img)
cv2.waitKey(0)
def show_unconfirmed_faces(args, svm_clf, knn_clf):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
unconfirmed = faces.get_unconfirmed(args.face)
if len(unconfirmed) == 0:
print('no newly predicted or unconfirmed faces found')
return False
files = faces.get_paths(unconfirmed, allow_duplicates=True)
i = 0
key = 0
while key != 27 and len(files) > 0:
if i < 0:
i = len(files) - 1
if i > len(files) - 1:
i = 0
str_count = str(i + 1) + ' / ' + str(len(files)) + ' #' + args.face + ': ' + str(len(unconfirmed))
print(str_count)
img_path = files[i]
opencvImage = cv2.imread(img_path)
height, width = opencvImage.shape[:2]
scale = 600.0 / float(height)
opencvImage = cv2.resize(opencvImage, (int(width * scale), int(height * scale)))
opencvImage_clean = opencvImage.copy()
main_idx = unconfirmed[i]
utils.draw_faces_on_image(faces, faces.dict_by_files[img_path], scale, opencvImage, main_idx)
utils.clicked_names, probs = utils.predict_face_svm(faces.get_face(main_idx).desc, svm_clf)
cv2.imshow("faces", opencvImage)
cv2.setMouseCallback("faces", utils.click_face, (opencvImage_clean, faces, scale, img_labels[img_path], svm_clf))
key = cv2.waitKey(0)
if len(utils.clicked_idx) == 0 and main_idx != -1:
utils.clicked_idx.append(main_idx)
utils.perform_key_action(args, key, faces, img_labels, utils.clicked_idx, utils.clicked_names, img_path, knn_clf, "")
utils.clicked_idx = []
if key == 46 or key == 47: # key '.' or key '/'
i += 1
elif key == 44: # key ','
i -= 1
elif key == 114: # key 'r'
i = random.randint(0, len(files) - 1)
utils.store_to_img_labels(faces, img_labels)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--imgs_root', type=str, required=True,
help="Path to images folder.")
args = parser.parse_args()
# dlib face detector
detector = dlib.get_frontal_face_detector()
sp = dlib.shape_predictor("models/shape_predictor_68_face_landmarks.dat")
facerec = dlib.face_recognition_model_v1("models/dlib_face_recognition_resnet_model_v1.dat")
# win = dlib.image_window()
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
for im in faces.dict_by_files:
for fi in faces.dict_by_files[im]:
face = faces.get_face(fi)
# if hasattr(face, 'shape_dlib68') and hasattr(face, 'desc_dlib68'):
# continue
img = dlib.load_rgb_image(face.path)
# locations: list of tuples (t,r,b,l)
# dlib rect: left: int, top: int, right: int, bottom: int
d = dlib.chip_details(dlib.rectangle(face.loc[3], face.loc[0], face.loc[1], face.loc[2]))
roi = dlib.extract_image_chip(np.array(img), d)
# idx tells which sub-detector was used, see https://github.com/davisking/dlib/blob/master/dlib/image_processing/frontal_face_detector.h
dets, scores, idx = detector.run(roi, 1, 0.3)
print(scores, idx)
if len(dets) == 0:
face.shape_dlib68 = None
face.desc_dlib68 = None
else:
shape = dlib.rectangle(dets[0].left()+face.loc[3], dets[0].top()+face.loc[0], dets[0].right()+face.loc[3], dets[0].bottom()+face.loc[0])
face.shape_dlib68 = sp(img, shape)
face.desc_dlib68 = np.array(facerec.compute_face_descriptor(img, face.shape_dlib68))
# win.clear_overlay()
# win.set_image(img)
# win.add_overlay(face.shape_dlib68)
# dlib.hit_enter_to_continue()
if not face.path in faces.changed_files:
faces.changed_files.append(face.path)
utils.store_to_img_labels(faces, img_labels)
def train_detectors(args):
X = []
y = []
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
for p in faces.dict_by_name:
if p not in ['unknown', 'deleted', 'detected', 'DELETED']:
for l in faces.dict_by_name[p]:
confirmed = faces.get_confirmed(l)
if confirmed not in [2]:
X.append(faces.get_desc(l))
y.append(p)
if len(X) == 0:
print('No faces found in database {}'.format(args.imgs_root))
return
# Determine how many neighbors to use for weighting in the KNN classifier
n_neighbors = int(round(math.sqrt(len(X))))
# n_neighbors = 100
print("Chose n_neighbors automatically:", n_neighbors)
# Create and train the KNN classifier
print("Training model with KNN ...")
knn_clf = neighbors.KNeighborsClassifier(n_neighbors=n_neighbors,
algorithm='auto',
weights='distance')
knn_clf.fit(X, y)
# Save the trained KNN classifier
with open(os.path.join(args.outdir, 'knn.clf'), 'wb') as f:
pickle.dump(knn_clf, f)
# train the svm
print("Training model with an SVM ...")
recognizer = SVC(C=1.0, kernel="linear", probability=True)
recognizer.fit(X, y)
# Save the trained SVM
with open(os.path.join(args.outdir, 'svm.clf'), 'wb') as f:
pickle.dump(recognizer, f)
print('Trained models with {} faces'.format(len(X)))
def show_faces_in_folder(args, svm_clf, knn_clf):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
files = faces.get_paths_from_folder(args.face)
i = 0
key = 0
while key != 27:
if i < 0:
i = len(files) - 1
if i > len(files) - 1:
i = 0
str_count = str(i + 1) + ' / ' + str(len(files)) + ' #' + args.face + ': ' + str(len(files))
print(str_count)
img_path = files[i]
opencvImage = cv2.imread(img_path)
height, width = opencvImage.shape[:2]
scale = 600.0 / float(height)
opencvImage = cv2.resize(opencvImage, (int(width * scale), int(height * scale)))
opencvImage_clean = opencvImage.copy()
utils.draw_faces_on_image(faces, faces.dict_by_folders[args.face][img_path], scale, opencvImage)
cv2.imshow("faces", opencvImage)
cv2.setMouseCallback("faces", utils.click_face, (opencvImage_clean, faces, scale, img_labels[img_path], svm_clf))
key = cv2.waitKey(0)
utils.perform_key_action(args, key, faces, img_labels, utils.clicked_idx, utils.clicked_names, img_path, knn_clf, "")
utils.clicked_idx = []
if key == 46 or key == 47: # key '.' or key '/'
i += 1
elif key == 44: # key ','
i -= 1
elif key == 114: # key 'r'
i = random.randint(0, len(files) - 1)
utils.store_to_img_labels(faces, img_labels)
def export_face_crops(args):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
sp = dlib.shape_predictor("models/shape_predictor_5_face_landmarks.dat")
for name in faces.dict_by_name:
if name == 'unknown' or name == 'deleted':
continue
face_dir = os.path.join(args.outdir, name)
if not os.path.isdir(face_dir):
utils.mkdir_p(face_dir)
print('Writing {}'.format(name))
for i, f in enumerate(faces.dict_by_name[name]):
face_path = os.path.join(face_dir, '{}_{:06d}.jpg'.format(name, i))
if not os.path.isfile(face_path) or args.overwrite:
if 0:
utils.save_face_crop(face_path, faces.get_face_path(f),
faces.get_loc(f))
else:
utils.save_face_crop_aligned(sp, face_path,
faces.get_face_path(f),
faces.get_loc(f))
# picture1_height = picture1.size[1]
# window.geometry("{}x{}+100+100".format(picture_width+picture1_width+10, picture_height))
# if i == 0:
# image_widget = tkinter.Label(window, image=tk_picture)
# image_widget1 = tkinter.Label(window, image=tk_picture1)
# else:
# image_widget.configure(image=tk_picture)
# image_widget1.configure(image=tk_picture1)
# image_widget.place(x=0, y=0, width=picture_width, height=picture_height)
# image_widget1.place(x=picture_width+10, y=0, width=picture1_width, height=picture1_height)
#
# # wait for events
# window.mainloop()
tmp_faces, img_labels = utils.load_img_labels("/Users/mhumenbe/Library/Mobile Documents/com~apple~CloudDocs/Fotos")
faces = utils.FACES(tmp_faces)
idxs = sorted(faces.dict_by_name['unknown'], key=lambda x: faces.get_face(x).timestamp, reverse=True)
files = faces.get_paths(idxs)
root = tk.Tk()
root.geometry("400x400")
def showimg(e):
n = lst.curselection()
fname = lst.get(n)
img = tk.PhotoImage(file=fname)
lab.config(image=img)
lab.image = img
print(fname)
def predict_faces(args, knn_clf, svm_clf):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
if not args.cls in faces.dict_by_name:
print('no faces found in this class')
return False
files = faces.get_paths(faces.dict_by_name[args.cls])
total_nr_files = len(files)
i = 0
key = 0
while key != 27 and i < total_nr_files:
str_count = str(i + 1) + ' / ' + str(total_nr_files)
print(str_count)
img_path = files[i]
opencvImage = cv2.imread(img_path)
height, width = opencvImage.shape[:2]
scale = 600.0 / float(height)
opencvImage = cv2.resize(opencvImage, (int(width * scale), int(height * scale)))
opencvImage_clean = opencvImage.copy()
idxs = sorted(faces.dict_by_name[args.cls], key=lambda x: faces.get_face(x).timestamp, reverse=True)
face_idxs = faces.get_face_idxs_by_name_and_file(args.cls, img_path, idxs)
for fi in face_idxs:
opencvImage = opencvImage_clean.copy()
face = faces.get_face(fi)
name = face.name
names, probs = utils.predict_face_svm(face.desc, svm_clf, print_top=True)
if name == "unknown" or name == "detected" and name != 'deleted' and name != 'DELETED':
if probs[0] >= 0.85:
name = names[0]
print('{} has high probability'.format(name))
else:
name_knn = utils.predict_knn(knn_clf, face.desc, n=7, thresh=0.3)
if name_knn != 'unknown':
name = name_knn
print('{} has majority in knn search'.format(name))
else:
name = 'unknown'
faces.rename(fi, name)
if name != args.cls and name != 'unknown' and name != 'deleted' and name != 'DELETED':
if args.confirm:
utils.show_face_crop(img_path, face.loc)
utils.draw_faces_on_image(faces, faces.dict_by_files[img_path], scale, opencvImage, fi)
cv2.putText(opencvImage, name, (20, opencvImage.shape[0] - 40), cv2.FONT_HERSHEY_SIMPLEX, 1,
(0, 0, 255), 1)
cv2.imshow("faces", opencvImage)
cv2.setMouseCallback("faces", utils.click_face,
(opencvImage_clean, faces, scale, img_labels[img_path], svm_clf))
key = cv2.waitKey(0)
if len(utils.clicked_idx) == 0:
utils.clicked_idx.append(fi)
utils.clicked_names = names
utils.perform_key_action(args, key, faces, img_labels, utils.clicked_idx, utils.clicked_names, img_path, knn_clf, "")
utils.clicked_idx = []
utils.clicked_names = []
if not args.cls in faces.dict_by_name:
print('no faces found in this class')
return False
if key == 46 or key == 47: # key '.' or key '/'
continue
elif key == 111: # 'o'
faces.rename(fi, name)
faces.set_confirmed(fi, 1)
else:
# move to new class
faces.rename(fi, name)
if name == 'unknown':
faces.set_confirmed(fi, 0)
else:
faces.set_confirmed(fi, 2)
i += 1
# files = faces.get_paths(faces.dict_by_name[args.cls])
utils.store_to_img_labels(faces, img_labels)
def tag_images(args):
# images = utils.get_images_in_dir_rec(args.input)
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
images = list(faces.dict_by_files.keys())
random.shuffle(images)
counter = 0
for img in images:
if counter >= 2000:
print('{} images processed'.format(counter))
return -1
pkl_path = img + '.pkl'
if os.path.isfile(pkl_path):
with open(pkl_path, 'rb') as fid:
img_label = pickle.load(fid)
else:
img_label = utils.IMG_LABELS(utils.get_timestamp(img))
img_label.path = img
# if hasattr(img_label, 'imagga'):
# rename_attribute(img_label, 'imagga', 'tags')
# with open(pkl_path, 'wb') as fid:
# pickle.dump(img_label, fid)
if not hasattr(img_label, 'categories'):
img_label.categories = []
if not hasattr(img_label, 'tags'):
img_label.tags = []
if len(img_label.tags) != 0 and len(img_label.categories) != 0:
print('{} already tagged'.format(img))
continue
print('tagging {}'.format(img))
response = requests.post(
'https://api.imagga.com/v2/uploads',
auth=(api_key, api_secret),
files={'image': open(img, 'rb')})
if response.ok:
image_upload_id = response.json()['result']['upload_id']
elif response.reason == 'Forbidden':
print(response.text)
return -1
else:
print(response.text)
continue
if len(img_label.tags) == 0:
response = requests.get(
'https://api.imagga.com/v2/tags?image_upload_id=%s' % image_upload_id,
auth=(api_key, api_secret))
if response.ok:
result_list = response.json()['result']['tags']
img_label.tags = [(t['tag']['en'], t['confidence']) for t in result_list if t['confidence'] > 0]
counter += 1
print('ok')
elif response.reason == 'Forbidden':
print(response.text)
return -1
else:
print(response.text)
continue
# if len(img_label.categories) == 0:
# categorizer_id = 'personal_photos'
# response = requests.get(
# 'https://api.imagga.com/v2/categories/%s?image_upload_id=%s' % (categorizer_id, image_upload_id),
# auth=(api_key, api_secret))
# if response.ok:
# result_list = response.json()['result']['categories']
# img_label.categories = [(t['name']['en'], t['confidence']) for t in result_list if t['confidence'] > 0]
# counter += 1
# else:
# print(response.text)
# return -1
with open(pkl_path, 'wb') as fid:
pickle.dump(img_label, fid)
def export_to_xmp_files(args):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
if len(faces.dict_by_files) == 0:
print('no faces loaded')
exit()
total_images = utils.get_images_in_dir_rec(args.imgs_root)
for f in total_images:
img_path = os.path.splitext(f)[0] + os.path.splitext(f)[1].lower()
if os.path.dirname(
img_path
) != args.mask_folder and args.mask_folder != None or img_path.lower(
).endswith('.png'):
continue
xmp_path = os.path.splitext(f)[0] + '.xmp'
if os.path.exists(xmp_path):
with open(xmp_path, 'r') as fptr:
strbuffer = fptr.read()
xmp = XMPMeta()
xmp.parse_from_str(strbuffer)
else:
xmpfile = XMPFiles(file_path=f, open_forupdate=True)
xmp = xmpfile.get_xmp()
# print(f)
kw_faces, kw_tags, kw_categories, kw_labels, kw_miscs = get_keywords(
xmp)
kw_tags = []
kw_categories = []
kw_miscs = []
if img_path in faces.dict_by_files:
names = faces.get_names(faces.dict_by_files[img_path])
# remove detected, deleted and unknown
unwanted_names = {'detected', 'deleted', 'unknown'}
names = [ele for ele in names if ele not in unwanted_names]
face_names = prepare_names(names, 'f ')
else:
# if not os.path.exists(xmp_path):
# continue
face_names = []
labels = []
if img_path in img_labels:
if len(img_labels[img_path].tags) != 0:
labels += prepare_names([
t[0].lower()
for t in img_labels[img_path].tags if t[1] >= 30
], 'l ')
if len(img_labels[img_path].categories) != 0:
labels += prepare_names(
[c[0].lower() for c in img_labels[img_path].categories],
'l ')
if hasattr(img_labels[img_path], 'gcloud_labels'):
labels += prepare_names(
[l.lower() for l in img_labels[img_path].gcloud_labels],
'l ')
if hasattr(img_labels[img_path], 'gcloud_objects'):
labels += prepare_names(
[l.lower() for l in img_labels[img_path].gcloud_objects],
'l ')
if hasattr(img_labels[img_path], 'gcloud_web'):
labels += prepare_names(
[l.lower() for l in img_labels[img_path].gcloud_web], 'l ')
if hasattr(img_labels[img_path], 'gcloud_landmarks'):
labels += prepare_names([
l.lower()
for l in img_labels[img_path].gcloud_landmarks[::2]
], 'l ')
labels = list(set(labels))
if sorted(face_names) != sorted(kw_faces) or sorted(labels) != sorted(
kw_labels) or not os.path.exists(xmp_path):
xmp.delete_property(consts.XMP_NS_DC, 'subject')
for face in face_names:
xmp.append_array_item(consts.XMP_NS_DC, 'subject', face, {
'prop_array_is_ordered': True,
'prop_value_is_array': True
})
for l in labels:
xmp.append_array_item(consts.XMP_NS_DC, 'subject', l, {
'prop_array_is_ordered': True,
'prop_value_is_array': True
})
print('exporting file: {}'.format(os.path.basename(xmp_path)))
with open(xmp_path, 'w') as fptr:
fptr.write(xmp.serialize_to_str(omit_packet_wrapper=True))
def export_to_csv(args):
tmp_faces, img_labels = utils.load_img_labels(args.imgs_root)
faces = utils.FACES(tmp_faces)
faces_csv_path = os.path.join(args.outdir, 'faces.csv')
faces_input_csv_path = os.path.join(args.outdir, 'faces_exiftool.csv')
if os.path.isfile(faces_input_csv_path):
files_faces_csv = utils.load_faces_from_keywords_csv(
faces_input_csv_path)
else:
files_faces_csv = None
with open(faces_csv_path, 'w+') as csvfile:
filewriter = csv.writer(csvfile, delimiter=',')
header = ['SourceFile', 'Keywords']
# header = ['SourceFile','Subject','XPKeywords','LastKeywordXMP','LastKeywordIPTC','UserComment']
filewriter.writerow(header)
if files_faces_csv != None:
for f in files_faces_csv:
full_path = os.path.join(args.imgs_root, f[2:])
real_image_path = getfile_sensitive(full_path)
relpath = './' + os.path.relpath(real_image_path,
args.imgs_root)
if os.path.splitext(relpath)[1].lower() in [
'.jpg', '.png'
] and full_path not in faces.dict_by_files:
row = [relpath, '-']
filewriter.writerow(row)
for e, f in enumerate(faces.dict_by_files):
print('{}/{}'.format(e, len(faces.dict_by_files)))
if os.path.dirname(
f) != args.mask_folder and args.mask_folder != None:
continue
real_image_path = getfile_sensitive(f)
relpath = './' + os.path.relpath(real_image_path, args.imgs_root)
relpath_lower = os.path.splitext(relpath)[0] + os.path.splitext(
relpath)[1].lower()
row = [relpath]
face_names = []
tmp_faces = []
if len(faces.dict_by_files[f]) == 1:
face_name = prepare_name(
faces.get_face(faces.dict_by_files[f][0]).name, 'f ')
tmp_faces.append(face_name)
if face_name not in ['f unknown', 'f deleted']:
face_names.append(face_name)
else:
str = ''
for i in faces.dict_by_files[f]:
face_name = prepare_name(faces.get_face(i).name, 'f ')
if not face_name in tmp_faces and face_name not in [
'f unknown', 'f deleted'
]:
str += face_name + ','
tmp_faces.append(face_name)
if str != '':
face_names.append(str[:-1])
if len(face_names) != 0:
if files_faces_csv != None:
if relpath in files_faces_csv:
if sorted(list(files_faces_csv[relpath].split(
', '))) == sorted(tmp_faces):
continue
elif relpath_lower in files_faces_csv:
if sorted(
list(files_faces_csv[relpath_lower].split(
', '))) == sorted(tmp_faces):
continue
row += face_names
filewriter.writerow(row)