Add show_points option to show_anns

[giswqs]

Summary

• Add show_points parameter to show_anns() that overlays point prompts as 5-pointed star markers on the annotation image, matching the style of show_points()
• Store point_coords and point_labels in predict_inst() so show_anns() can automatically display them without re-specifying
• Supports custom colors, sizes, and explicit point overrides via parameters

Usage

# Automatic - uses points from the last predict_inst() / generate_masks_by_points() call
sam.show_anns(show_points=True)

# Explicit points
sam.show_anns(show_points=True, point_coords=[[520, 375]], point_labels=[1])

Test plan

• Verified show_anns(show_points=True) renders star markers after generate_masks_by_points()
• Verified foreground (dark green) and background (red) point colors
• Verified white edge outline matches show_points() style
• Verified show_anns() without show_points is unchanged (backward compatible)
• Pre-commit checks pass

[github-actions]

🚀 Deployed on https://69be05c58945e45fc43fe427--opengeos.netlify.app

[Copilot]

Pull request overview

This PR extends SamGeo3.show_anns() to optionally overlay point prompts (as star markers) on the rendered annotation image, using either explicitly provided points or the most recently used prompts stored from interactive prediction.

Changes:

• Added show_points and related point-visualization parameters to show_anns() to draw star markers with configurable colors/sizing.
• Persisted point_coords/point_labels in predict_inst() for reuse by show_anns(show_points=True).
• Updated show_anns() docstring to describe the new point overlay behavior and parameters.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

When show_points=True, point_labels is converted to a numpy array and then zipped with coords_arr. If point_labels length doesn’t match point_coords, zip() will silently drop extra points; if point_labels is passed as a scalar (common elsewhere in this repo), iterating a 0-d numpy array will raise. Consider normalizing/broadcasting labels (e.g., scalar or length-1 -> repeat) and raising a clear ValueError when lengths still don’t match.

Suggested change

	if labels is None:
	labels = [1] * len(coords_arr)
	labels_arr = np.array(labels)
	n_points = len(coords_arr)
	# Normalize/broadcast labels to match the number of points.
	if labels is None:
	# Default: all points are treated as foreground prompts.
	labels_arr = np.ones(n_points, dtype=int)
	else:
	labels_arr = np.array(labels)
	# Handle scalar labels (0-D array) by broadcasting.
	if labels_arr.ndim == 0:
	labels_arr = np.full(n_points, int(labels_arr), dtype=int)
	else:
	# Ensure 1-D and handle single-element sequences.
	labels_arr = labels_arr.reshape(-1)
	if labels_arr.size == 1 and n_points > 1:
	labels_arr = np.full(n_points, int(labels_arr[0]), dtype=int)
	elif labels_arr.size != n_points:
	raise ValueError(
	f"Number of point_labels ({labels_arr.size}) does not match "
	f"number of point_coords ({n_points})."
	)

[Copilot]

_star_pts is defined inside the per-point loop, so it is re-created for every point rendered. Moving this helper outside the loop (or precomputing a template and translating it per point) would reduce overhead and keep the rendering logic simpler.

Suggested change

	for pt, lbl in zip(coords_arr, labels_arr):
	x, y = int(pt[0]), int(pt[1])
	color = foreground_color if lbl == 1 else background_color
	# Build a 5-pointed star polygon
	def _star_pts(cx, cy, r_outer, r_inner, n=5):
	pts = []
	for i in range(2 * n):
	r = r_outer if i % 2 == 0 else r_inner
	angle = np.pi / 2 + i * np.pi / n
	pts.append(
	[
	int(cx + r * np.cos(angle)),
	int(cy - r * np.sin(angle)),
	]
	)
	return np.array(pts, dtype=np.int32)
	# Build a 5-pointed star polygon once per call, reuse for each point
	def _star_pts(cx, cy, r_outer, r_inner, n=5):
	pts = []
	for i in range(2 * n):
	r = r_outer if i % 2 == 0 else r_inner
	angle = np.pi / 2 + i * np.pi / n
	pts.append(
	[
	int(cx + r * np.cos(angle)),
	int(cy - r * np.sin(angle)),
	]
	)
	return np.array(pts, dtype=np.int32)
	for pt, lbl in zip(coords_arr, labels_arr):
	x, y = int(pt[0]), int(pt[1])
	color = foreground_color if lbl == 1 else background_color

[Copilot]

point_size is documented as “Size of star markers in pixels”, but it’s used as the outer radius when computing the star vertices (the marker ends up roughly ~2*point_size across). Please clarify the docstring (e.g., outer radius vs diameter) or rename the parameter to avoid confusion for callers tuning marker size.

Suggested change

	point_size (int): Size of star markers in pixels. Defaults to 15.
	point_size (int): Outer radius of star markers in pixels (marker
	diameter is roughly 2 * point_size). Defaults to 15.

[Copilot]

predict_inst() now stores self.point_coords/self.point_labels for later visualization, but it doesn’t normalize or validate point_labels relative to point_coords before storing. If a caller provides a scalar label or a mismatched length (both patterns supported in other modules), show_anns(show_points=True) can later fail or render incorrectly. Consider broadcasting/validating labels here before saving them (or saving a normalized Nx2 / length-N representation only).

Suggested change

	self.point_coords = (
	point_coords.tolist()
	if isinstance(point_coords, np.ndarray)
	else point_coords
	)
	self.point_labels = (
	point_labels.tolist()
	if isinstance(point_labels, np.ndarray)
	else point_labels
	)
	# Normalize point coordinates to a Python list and infer number of points.
	if isinstance(point_coords, np.ndarray):
	coords_array = point_coords
	coords_list = point_coords.tolist()
	else:
	coords_array = np.asarray(point_coords)
	coords_list = point_coords
	# Try to determine how many points we have.
	n_points: int
	if coords_array.ndim == 2:
	# Typical case: (N, 2)
	n_points = int(coords_array.shape[0])
	elif coords_array.ndim == 1:
	# Treat a flat 1D coordinate as a single point (e.g., [x, y]).
	n_points = 1
	else:
	# Fallback: rely on Python's len().
	try:
	n_points = len(coords_list)
	except TypeError:
	n_points = 1
	# Normalize point labels to match the number of points.
	labels_norm = None
	if point_labels is not None:
	if isinstance(point_labels, np.ndarray):
	labels_raw = point_labels.tolist()
	else:
	labels_raw = point_labels
	# Handle scalar label by broadcasting to all points.
	if isinstance(
	labels_raw,
	(int, float, bool, np.integer, np.floating),
	):
	labels_norm = [labels_raw] * max(n_points, 0)
	else:
	# Ensure we have a list for length checks/broadcasting.
	labels_list = list(labels_raw)
	if n_points <= 1:
	# Single point: take the first label if available.
	if labels_list:
	labels_norm = [labels_list[0]]
	else:
	labels_norm = []
	else:
	if len(labels_list) == 1:
	# Broadcast single label to all points.
	labels_norm = labels_list * n_points
	elif len(labels_list) == n_points:
	labels_norm = labels_list
	else:
	raise ValueError(
	f"Length of point_labels ({len(labels_list)}) does not match "
	f"number of point_coords ({n_points})."
	)
	self.point_coords = coords_list
	self.point_labels = labels_norm