def city_name(self) -> str:
"""get city name of the current log
Returns:
city_name: city name of the current log, either 'PIT' or 'MIA'
"""
return read_city_name(
os.path.join(self.root_dir, self.current_log, "city_info.json"))
def get_city_name(self, log_id: str) -> str:
"""
Args:
log_id: str
Returns:
city_name: str
"""
city_info_fpath = f"{self.data_dir}/{log_id}/city_info.json"
city_name = read_city_name(city_info_fpath)
assert isinstance(city_name, str)
return city_name
def get_city_name(self, log_id: str) -> str:
"""Return the name of the city where the log of interest was captured.
Args:
log_id: str
Returns:
city_name: str
"""
city_info_fpath = f"{self.data_dir}/{log_id}/city_info.json"
city_name = read_city_name(city_info_fpath)
assert isinstance(city_name, str)
return city_name
def accumulate(
dt_root_fpath: Path, gt_fpath: Path, cfg: DetectionCfg,
avm: Optional[ArgoverseMap]
) -> Tuple[DefaultDict[str, np.ndarray], DefaultDict[str, int]]:
"""Accumulate the true/false positives (boolean flags) and true positive errors for each class.
Args:
dt_root_fpath: Detections root folder file path.
gt_fpath: Ground truth file path.
cfg: Detection configuration.
Returns:
cls_to_accum: Class to accumulated statistics dictionary of shape |C| -> (N, K + S) where C
is the number of detection classes, K is the number of true positive thresholds used for
AP computation, and S is the number of true positive errors.
cls_to_ninst: Mapping of shape |C| -> (1,) the class names to the number of instances in the ground
truth dataset.
"""
log_id = gt_fpath.parents[1].stem
logger.info(f"log_id = {log_id}")
ts = int(gt_fpath.stem.split("_")[-1])
dt_fpath = dt_root_fpath / f"{log_id}/per_sweep_annotations_amodal/" f"tracked_object_labels_{ts}.json"
dts = np.array(read_label(str(dt_fpath)))
gts = np.array(read_label(str(gt_fpath)))
if cfg.eval_only_roi_instances and avm is not None:
# go up 3 levels, because hierarchy is as follows:
# {gt_root_fpath}/{log_id}/per_sweep_annotations_amodal/{gt_root_fname}
gt_root_fpath = Path(gt_fpath).parents[2]
city_SE3_egovehicle = get_city_SE3_egovehicle_at_sensor_t(
ts, str(gt_root_fpath), log_id)
if city_SE3_egovehicle is not None:
log_city_name = read_city_name(
os.path.join(gt_root_fpath, log_id, "city_info.json"))
dts = filter_objs_to_roi(dts, avm, city_SE3_egovehicle,
log_city_name)
gts = filter_objs_to_roi(gts, avm, city_SE3_egovehicle,
log_city_name)
cls_to_accum = defaultdict(list)
cls_to_ninst = defaultdict(int)
for class_name in cfg.dt_classes:
dt_filtered = filter_instances(
dts,
class_name,
filter_metric=cfg.dt_metric,
max_detection_range=cfg.max_dt_range,
)
gt_filtered = filter_instances(
gts,
class_name,
filter_metric=cfg.dt_metric,
max_detection_range=cfg.max_dt_range,
)
gt_filtered = remove_duplicate_instances(gt_filtered, cfg)
logger.info(f"{dt_filtered.shape[0]} detections")
logger.info(f"{gt_filtered.shape[0]} ground truth")
if dt_filtered.shape[0] > 0:
ranked_detections, scores = rank(dt_filtered)
metrics = assign(ranked_detections, gt_filtered, cfg)
cls_to_accum[class_name] = np.hstack((metrics, scores))
cls_to_ninst[class_name] = gt_filtered.shape[0]
return cls_to_accum, cls_to_ninst