def _convert_bbox(self, delta, point):
delta = delta.permute(1, 2, 3, 0).contiguous().view(4, -1)
delta = delta.detach().cpu().numpy()
delta[0, :] = point[:, 0] - delta[0, :]
delta[1, :] = point[:, 1] - delta[1, :]
delta[2, :] = point[:, 0] + delta[2, :]
delta[3, :] = point[:, 1] + delta[3, :]
delta[0, :], delta[1, :], delta[2, :], delta[3, :] = corner2center(delta)
return delta
def __call__(self, target, size, neg=False):
# -1 ignore; 0 negative; 1 positive
# 使用-1初始化cls, 之后负采样会设置0, 正采样会设置1
cls = -1 * np.ones((size, size), dtype=np.int64)
delta = np.zeros((4, size, size), dtype=np.float32)
tcx, tcy, tw, th = corner2center(target)
points = self.points.points
def select(position, keep_num=16):
num = position[0].shape[0]
if num <= keep_num:
return position, num
slt = np.arange(num)
np.random.shuffle(slt)
slt = slt[:keep_num]
return tuple(p[slt] for p in position), keep_num
if neg:
neg = np.where(
np.square(tcx - points[0]) / np.square(tw / 4) +
np.square(tcy - points[1]) / np.square(th / 4) < 1)
neg, neg_num = select(neg, cfg.TRAIN.NEG_NUM)
cls[neg] = 0
return cls, delta
delta[0] = points[0] - target[0]
delta[1] = points[1] - target[1]
delta[2] = target[2] - points[0]
delta[3] = target[3] - points[1]
# 椭圆标签
# ellipse label
pos = np.where(
np.square(tcx - points[0]) / np.square(tw / 4) +
np.square(tcy - points[1]) / np.square(th / 4) < 1)
neg = np.where(
np.square(tcx - points[0]) / np.square(tw / 2) +
np.square(tcy - points[1]) / np.square(th / 2) > 1)
# sampling
pos, pos_num = select(pos, cfg.TRAIN.POS_NUM)
neg, neg_num = select(neg, cfg.TRAIN.TOTAL_NUM - cfg.TRAIN.POS_NUM)
cls[pos] = 1
cls[neg] = 0
return cls, delta
def main():
# load config
cfg.merge_from_file(args.config)
cur_dir = os.path.dirname(os.path.realpath(__file__))
# create model
model = ModelBuilder()
# load model
model = load_pretrain(model, args.snapshot).cuda().eval()
# build tracker
tracker = build_tracker(model)
handle = vot.VOT("rectangle")
region = handle.region()
imagefile = handle.frame()
if not imagefile:
sys.exit(0)
img = cv2.imread(imagefile)
left = max(region.x, 0)
top = max(region.y, 0)
right = min(region.x + region.width, img.shape[1] - 1)
bottom = min(region.y + region.height, img.shape[0] - 1)
cx, cy, w, h = corner2center(Corner(left, top, right, bottom))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
tracker.init(img, gt_bbox_)
while True:
imagefile = handle.frame()
if not imagefile:
break
image = cv2.imread(imagefile)
outputs = tracker.track(image)
pred_bbox = outputs['bbox']
conf = outputs['best_score']
handle.report(vot.Rectangle(*pred_bbox), conf)
def _shift_scale_aug(self, image, bbox, crop_bbox, size):
im_h, im_w = image.shape[:2]
# adjust crop bounding box
crop_bbox_center = corner2center(crop_bbox)
if self.scale:
scale_x = (1.0 + Augmentation.random() * self.scale)
scale_y = (1.0 + Augmentation.random() * self.scale)
h, w = crop_bbox_center.h, crop_bbox_center.w
scale_x = min(scale_x, float(im_w) / w)
scale_y = min(scale_y, float(im_h) / h)
crop_bbox_center = Center(crop_bbox_center.x, crop_bbox_center.y,
crop_bbox_center.w * scale_x,
crop_bbox_center.h * scale_y)
crop_bbox = center2corner(crop_bbox_center)
if self.shift:
sx = Augmentation.random() * self.shift
sy = Augmentation.random() * self.shift
x1, y1, x2, y2 = crop_bbox
sx = max(-x1, min(im_w - 1 - x2, sx))
sy = max(-y1, min(im_h - 1 - y2, sy))
crop_bbox = Corner(x1 + sx, y1 + sy, x2 + sx, y2 + sy)
# adjust target bounding box
x1, y1 = crop_bbox.x1, crop_bbox.y1
bbox = Corner(bbox.x1 - x1, bbox.y1 - y1, bbox.x2 - x1, bbox.y2 - y1)
if self.scale:
bbox = Corner(bbox.x1 / scale_x, bbox.y1 / scale_y,
bbox.x2 / scale_x, bbox.y2 / scale_y)
image = self._crop_roi(image, crop_bbox, size)
return image, bbox
