def try_segment_objects(
img_files: Sequence[Union[str, PathLike]],
application: Application,
model: Optional["Model"] = None,
channelwise_minmax: bool = False,
channelwise_zscore: bool = False,
channel_groups: Optional[np.ndarray] = None,
aggr_func: Callable[[np.ndarray], np.ndarray] = np.mean,
**predict_kwargs,
) -> Generator[Tuple[Path, np.ndarray], None, None]:
app, predict = application.value(model=model)
for img_file in img_files:
try:
img = io.read_image(img_file)
if channelwise_minmax:
channel_mins = np.nanmin(img, axis=(1, 2), keepdims=True)
channel_maxs = np.nanmax(img, axis=(1, 2), keepdims=True)
img = (img - channel_mins) / (channel_maxs - channel_mins)
if channelwise_zscore:
channel_means = np.nanmean(img, axis=(1, 2), keepdims=True)
channel_stds = np.nanstd(img, axis=(1, 2), keepdims=True)
img = (img - channel_means) / channel_stds
if channel_groups is not None:
img = np.stack(
[
aggr_func(img[channel_groups == channel_group], axis=0)
for channel_group in np.unique(channel_groups)
if not np.isnan(channel_group)
]
)
mask = predict(img, **predict_kwargs)
yield Path(img_file), mask
del img, mask
except:
_logger.exception(f"Error segmenting objects in {img_file}")