def __init__(self, img, truth, is_training, batcn_norm_decay=0.997):
self.img = img
self.truth = truth
self.is_training = is_training
self.batch_norm_decay = batcn_norm_decay
self.img_shape = tf.shape(self.img)
backbone = Network()
if is_training:
self.head, self.l2_loss = backbone.inference(self.is_training, self.img)
else:
self.head = backbone.inference(self.is_training, self.img)
def __init__(self, img, age_labels, age_vector, is_training, batcn_norm_decay=0.997):
self.img = img
self.age_labels = age_labels
self.age_vector = age_vector
self.is_training = is_training
self.batch_norm_decay = batcn_norm_decay
self.img_shape = tf.shape(self.img)
backbone = Network()
if is_training:
self.feats, self.pred, self.l1_loss = backbone.inference(self.is_training, self.img)
else:
self.feats, self.pred = backbone.inference(self.is_training, self.img)
if args.cfg:
cfg_from_file(args.cfg)
net = Network()
checkpoint = torch.load(osp.abspath(args.checkpoint))
net.load_state_dict(checkpoint["model"])
logging.info("Loaded checkpoint from: %s" % args.checkpoint)
net.eval()
device = torch.device("cuda:%s" % args.gpu if args.gpu != -1 else "cpu")
net.to(device)
# Extract feature of the query person
query_img = cv2.imread("imgs/query.jpg")
query_roi = np.array([0, 0, 466, 943]) # [x1, y1, x2, y2]
query_feat = net.inference(query_img, query_roi).view(-1, 1)
# Get gallery images
gallery_imgs = sorted(glob("imgs/gallery*.jpg"))
for gallery_img in gallery_imgs:
logging.info("Detecting %s" % gallery_img)
detections, features = net.inference(cv2.imread(gallery_img))
# Compute pairwise cosine similarities,
# which equals to inner-products, as features are already L2-normed
similarities = features.mm(query_feat).squeeze()
# Visualize the results
visualize_result(gallery_img, detections, similarities)
# which equals to inner-products, as features are already L2-normed
similarities = features.mm(query_feat).squeeze()
# Visualize the results
visualize_result(gallery_img, detections, similarities)
"""
#### For video file
cap = cv2.VideoCapture("data/dataset/video/handmovewhite.mp4")
#### For video camera
#cap = cv2.VideoCapture(0)
query_img = cv2.imread("data/dataset/video/query.png")
query_roi = np.array([0, 0, 313, 733])
query_feat = net.inference(query_img, query_roi).view(-1, 1)
scale_percent = 50
width = int(cap.get(3) * scale_percent / 100)
height = int(cap.get(4) * scale_percent / 100)
dim = (width, height)
output_video = cv2.VideoWriter("output/output_video.avi", cv2.VideoWriter_fourcc('M','J','P','G'), 20.0, (width, height))
no_frame = 1
while cap.isOpened():
ret, frame = cap.read()
if not ret:
print("Can't receive frame (stream end?). Exiting ...")
