def __init__(self, conf, verbose=True):
self.device = conf.device
self.threshold = conf.threshold
try:
if conf.use_mobilfacenet:
self.model = MobileFaceNet(conf.embedding_size).to(conf.device)
# download the default weigth
file_name = os.path.join(
os.path.dirname(os.path.realpath(__file__)),
'mobilenet.pth')
weight_path = os.path.join(os.path.dirname(file_name),
'weights/mobilenet.pth')
if os.path.isfile(weight_path) == False:
os.makedirs(os.path.split(weight_path)[0], exist_ok=True)
download_weight(
link=
'https://drive.google.com/uc?export=download&id=1W9nM7LE6zUKQ4tncL6OnBn-aXNyiRPNH',
file_name=file_name,
verbose=verbose)
os.rename(file_name, weight_path)
self.model.load_state_dict(
torch.load(weight_path, map_location=conf.device))
else:
self.model = Backbone(conf.net_depth, conf.drop_ratio,
conf.net_mode).to(conf.device)
# download the default weigth
file_name = os.path.join(
os.path.dirname(os.path.realpath(__file__)), 'ir_se50.pth')
weight_path = os.path.join(os.path.dirname(file_name),
'weights/ir_se50.pth')
if os.path.isfile(weight_path) == False:
os.makedirs(os.path.split(weight_path)[0], exist_ok=True)
download_weight(
link=
'https://www.dropbox.com/s/a570ucg6a5z22zf/ir_se50.pth?dl=1',
file_name=file_name,
verbose=verbose)
os.rename(file_name, weight_path)
self.model.load_state_dict(
torch.load(weight_path, map_location=conf.device))
except IOError as e:
exit(
f'from FaceRecognizer Exit: the weight does not exist,'
f' \n download and putting up in "{conf.work_path}" folder \n {e}'
)
self.model.eval()
torch.set_grad_enabled(False)
def __init__(self, name='mobilenet', weight_path=None, device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu"), confidence_threshold=0.99,
top_k=5000, nms_threshold=0.4, keep_top_k=750, face_size=(112, 112)):
"""
RetinaFace Detector with 5points landmarks
Args:
name: name of backbone (resnet, mobilenet, slim, rfb)
weight_path: path of network weight
device: running device (cuda, cpu)
face_size: final face size
face_padding: padding for bounding boxes
"""
model, cfg = None, None
if name == 'mobilenet':
cfg = cfg_mnet
model = RetinaFace(cfg=cfg, phase='test')
elif name == 'resnet':
cfg = cfg_re50
model = RetinaFace(cfg=cfg, phase='test')
else:
exit('from FaceDetector Exit: model not support \n just(mobilenet, resnet)')
# download the default weigth
if weight_path is None:
file_name = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'mobilenet.pth')
weight_path = os.path.join(os.path.dirname(file_name), 'weights/mobilenet.pth')
if os.path.isfile(weight_path) == False:
print('from FaceDetector: download defualt weight started')
os.makedirs(os.path.split(weight_path)[0], exist_ok=True)
download_weight(link='https://drive.google.com/uc?export=download&id=15zP8BP-5IvWXWZoYTNdvUJUiBqZ1hxu1', file_name=file_name)
os.rename(file_name, weight_path)
# setting for model
model.load_state_dict(torch.load(weight_path, map_location=device))
model.to(device).eval()
self.model = model
self.device = device
self.cfg = cfg
# setting for face detection
self.thresh = confidence_threshold
self.top_k = top_k
self.nms_thresh = nms_threshold
self.keep_top_k = keep_top_k
# setting for face align
self.trans = transform.SimilarityTransform()
self.out_size = face_size
self.ref_pts = get_reference_facial_points(output_size=face_size)
print('from FaceDetector: weights loaded')
return
def __init__(self, name='full', weight_path=None, device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu")):
"""
Age and gender Detector
:param name: name of backbone (full or tiny)
:param device: model run in cpu or gpu (cuda, cpu)
:param weight_path: path of network weight
Notice: image size must be 112x112
but cun run with 224x224
Method detect:
:param faces: 4d tensor of face for example size(1, 3, 112, 112)
:returns genders list and ages list
"""
if name == 'tiny':
model = ShuffleneTiny()
elif name == 'full':
model = ShuffleneFull()
else:
exit('from AgeGender Detector: model dose not support just(tiny, full)')
# download the default weigth
if weight_path is None:
file_name = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'ShufflenetFull.pth')
weight_path = os.path.join(os.path.dirname(file_name), 'weights/ShufflenetFull.pth')
if os.path.isfile(weight_path) == False:
print('from AgeGenderEstimator: download defualt weight started')
os.makedirs(os.path.split(weight_path)[0], exist_ok=True)
download_weight(link='https://drive.google.com/uc?export=download&id=1rnOZo46RjGZYrUb6Wup6sSOP37ol5I9E', file_name=file_name)
os.rename(file_name, weight_path)
model.load_state_dict(torch.load(weight_path, map_location=device))
model.to(device).eval()
self.model = model
self.device = device
print('from AgeGenderEstimator: weights loaded')
def __init__(self, name='densnet121', weight_path=None, device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu")):
"""
Residual Masking Emotion Detector from a list of labels
:param name: name of backbone of networks (resnet34, densenet121)
:param device: model run in cpu or gpu (cuda, cpu)
:param weight_path: path of network weight
Notice: image size must be 224x224
Method detect_emotion:
:param faces: 4d tensor of face for example size(1, 3, 224, 224)
:returns emotions list and probability of emotions
"""
self.device = device
self.model = None
if name == 'resnet34':
self.model = resnet34()
elif name == 'densnet121':
self.model = densenet121()
else:
exit('from EmotionDetector: Network does not support!! \n just(resnet34, densnet121)')
# download the default weigth
if weight_path is None:
file_name = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'densnet121.pth')
weight_path = os.path.join(os.path.dirname(file_name), 'weights/densnet121.pth')
if os.path.isfile(weight_path) == False:
print('from EmotionDetector: download defualt weight started')
os.makedirs(os.path.split(weight_path)[0], exist_ok=True)
download_weight(link='https://drive.google.com/uc?export=download&id=1G3VsfgiQb16VyFnOwEVDgm2g8-9qN0-9', file_name=file_name)
os.rename(file_name, weight_path)
self.model.load_state_dict(torch.load(weight_path, map_location=device))
self.model.to(device).eval()
print('from EmotionDetector: weights loaded')
def __init__(self, conf, inference=True):
if conf.use_mobilfacenet:
self.model = MobileFaceNet(conf.embedding_size).to(conf.device)
else:
self.model = Backbone(conf.net_depth, conf.drop_ratio,
conf.net_mode).to(conf.device)
if not inference:
self.milestones = conf.milestones
self.loader, self.class_num = get_train_loader(conf)
self.step = 0
self.head = Arcface(embedding_size=conf.embedding_size,
classnum=self.class_num).to(conf.device)
print('two model heads generated')
paras_only_bn, paras_wo_bn = separate_bn_paras(self.model)
if conf.use_mobilfacenet:
self.optimizer = optim.SGD(
[{
'params': paras_wo_bn[:-1],
'weight_decay': 4e-5
}, {
'params': [paras_wo_bn[-1]] + [self.head.kernel],
'weight_decay': 4e-4
}, {
'params': paras_only_bn
}],
lr=conf.lr,
momentum=conf.momentum)
else:
self.optimizer = optim.SGD(
[{
'params': paras_wo_bn + [self.head.kernel],
'weight_decay': 5e-4
}, {
'params': paras_only_bn
}],
lr=conf.lr,
momentum=conf.momentum)
print(self.optimizer)
print('optimizers generated')
self.board_loss_every = len(self.loader) // 100
self.evaluate_every = len(self.loader) // 10
self.save_every = len(self.loader) // 5
self.agedb_30, self.cfp_fp, self.lfw, self.agedb_30_issame, self.cfp_fp_issame, self.lfw_issame = get_val_data(
self.loader.dataset.root.parent)
else:
self.threshold = conf.threshold
try:
if conf.use_mobilfacenet:
# download the default weigth
file_name = os.path.join(
os.path.dirname(os.path.realpath(__file__)),
'mobilenet.pth')
weight_path = os.path.join(os.path.dirname(file_name),
'weights/mobilenet.pth')
if os.path.isfile(weight_path) == False:
print(
'from FaceRecognizer: download defualt weight started'
)
os.makedirs(os.path.split(weight_path)[0],
exist_ok=True)
download_weight(
link=
'https://drive.google.com/uc?export=download&id=1W9nM7LE6zUKQ4tncL6OnBn-aXNyiRPNH',
file_name=file_name)
os.rename(file_name, weight_path)
self.model.load_state_dict(
torch.load(weight_path, map_location=conf.device))
print('from FaceRecognizer: MobileFaceNet Loaded')
else:
self.model.load_state_dict(
torch.load(f'{conf.work_path}/ir_se50.pth'))
print('from FaceRecognizer: IR_SE_50 Loaded')
except IOError as e:
exit(
f'from FaceRecognizer Exit: the weight does not exist,'
f' \n download and putting up in "{conf.work_path}" folder \n {e}'
)