Translation from GUI settings to Python API not always completely clear

[oanegros]

Hi,

I'm having an issue with the translation of our pipeline as we're executing it in the plantseg 2 GUI to a python script. Namely, I can't find the equivalent function call for 'Remove objects with low foreground probability'. Does this exist, and which function is this?

Also, some other parts of this are still somewhat untransparent, such as the naming renaming agglomeration modes between camelcase and snakecase in GUI vs Python.

Thanks :D

[qin-yu]

Hey Oane,

GUI widgets correspond to tasks which wraps functionals for resource management. In IDEs or on GitHub, you can usually click into a call and see what is underneath.

plant-seg/plantseg/functionals/dataprocessing/advanced_dataprocessing.py

Lines 305 to 357 in f7e8e4b

	def remove_false_positives_by_foreground_probability(
	segmentation: np.ndarray,
	foreground: np.ndarray,
	threshold: float,
	) -> np.ndarray:
	"""
	Removes false positive regions in a segmentation based on a foreground probability map.
	1. Labels are not preserved.
	2. If the mean(an instance * its own probability region) < threshold, it is removed.
	Args:
	segmentation (np.ndarray): Segmentation array where each unique non-zero value indicates a distinct region.
	foreground (np.ndarray): Foreground probability map of the same shape as `segmentation`.
	threshold (float): Probability threshold below which regions are considered false positives.
	Returns:
	np.ndarray: Segmentation array with false positives removed.
	"""
	# TODO: make a channel for removed regions for easier inspection
	# TODO: use `relabel_sequential` to recover the original labels
	if segmentation.shape != foreground.shape:
	raise ValueError("Segmentation and probability map must have the same shape.")
	if foreground.max() > 1:
	raise ValueError("Foreground must be a probability map with values in [0, 1].")
	instances, _, _ = relabel_sequential(segmentation) # The label 0 is assumed to denote the bg and is never remapped.
	regions = regionprops(instances) # Labels with value 0 are ignored.
	pixel_count = np.zeros(len(regions) + 1)
	pixel_value = np.zeros(len(regions) + 1)
	pixel_count[0] = 1 # Avoid division by zero: pixel_count[0] and pixel_value[0] are fixed throughout.
	for region in tqdm.tqdm(regions):
	bbox = region.bbox
	if instances.ndim == 3:
	slices = (slice(bbox[0], bbox[3]), slice(bbox[1], bbox[4]), slice(bbox[2], bbox[5]))
	else:
	slices = (slice(bbox[0], bbox[2]), slice(bbox[1], bbox[3]))
	region_mask = instances[slices] == region.label
	prob = foreground[slices]
	pixel_count[region.label] = region.area
	pixel_value[region.label] = (region_mask * prob).sum()
	likelihood = pixel_value / pixel_count
	to_remove = likelihood < threshold
	instances[np.isin(instances, np.nonzero(to_remove)[0])] = 0
	instances, _, _ = relabel_sequential(instances)
	return instances