def __call__(self, data: Dict) -> bool:
if data is None:
return True
im, anno = data["image"], data["anno"]
if self._hyper_params["target_type"] == "bbox":
bbox = xyxy2xywh(anno)
elif self._hyper_params["target_type"] == "mask":
bbox = cv2.boundingRect(anno)
else:
logger.error("unspported target type {} in filter".format(
self._hyper_params["target_type"]))
exit()
filter_flag = filter_unreasonable_training_boxes(
im, bbox, self._hyper_params)
return filter_flag
dev = torch.device(parsed_args.device)
pipeline.set_device(dev)
if __name__ == "__main__":
seq = dataset[parsed_args.sequence_index]
template_frame = {k: seq[k][parsed_args.template_frame] for k in seq}
template_frame['image'] = load_image(template_frame['image'])
search_frame = {k: seq[k][parsed_args.search_frame] for k in seq}
search_frame['image'] = load_image(search_frame['image'])
im = template_frame['image']
bbox = template_frame['anno']
rect = xyxy2xywh(bbox)
pipeline.init(im, rect)
bbox = tuple(map(int, bbox))
cv2.rectangle(im,
bbox[:2],
bbox[2:],
color["target"],
thickness=bbox_thickness)
cv2.rectangle(im, (0, 0), (im.shape[1] - 1, im.shape[0] - 1),
color["border"],
thickness=10)
im = cv2.resize(im, (0, 0), fx=resize_factor, fy=resize_factor)
im = cv2.putText(im, "template frame", (20, 20), cv2.FONT_HERSHEY_COMPLEX,
font_size, color["target"], font_width)
logger.info("Load experiment configuration at: %s" % exp_cfg_path)
# resolve config
root_cfg = root_cfg.test
task, task_cfg = specify_task(root_cfg)
task_cfg.freeze()
# build model
model = model_builder.build(task, task_cfg.model)
# build pipeline
pipeline = pipeline_builder.build(task, task_cfg.pipeline, model)
dev = torch.device(parsed_args.device)
pipeline.set_device(dev)
if __name__ == "__main__":
rect = xyxy2xywh(bbox)
pipeline.init(im, rect)
im_size = np.array((im.shape[1], im.shape[0]), dtype=np.float)
crop_pos = np.array([parsed_args.shift_x, parsed_args.shift_y])
im_shift = get_subwindow(im, im_size * crop_pos, im_size, im_size)
rect_pred = pipeline.update(im_shift)
bbox_pred = xywh2xyxy(rect_pred)
bbox_pred = tuple(map(int, bbox_pred))
im_ = im_shift
cv2.rectangle(im_, bbox[:2], bbox[2:], color["target"])
cv2.rectangle(im_, bbox_pred[:2], bbox_pred[2:], color["pred"])
cv2.rectangle(im_, (0, 0), (im.shape[1] - 1, im.shape[0] - 1),
color["border"],