Plane detection

Instantiate a plane detector

limap.image.groups.planelib.get_plane_detector(method: str, plane_options: DetectorOptions): Get a plane detector

class limap.image.groups.planelib.DetectorOptions(base_options: limap.image.groups.planelib.base_plane_detector.BasePlaneDetectorOptions = <factory>, pxwplanar_options: limap.image.groups.planelib.register_plane_detector.PxwPlanarOptions = <factory>)

__init__(base_options: BasePlaneDetectorOptions = <factory>, pxwplanar_options: PxwPlanarOptions = <factory>) → None

base_options: BasePlaneDetectorOptions

pxwplanar_options: PxwPlanarOptions

class limap.image.groups.planelib.PxwPlanarOptions(model_path: str = 'alpayozkan/pxwplanar-moge2-planarity', device: str = 'cuda', num_tokens: int = 1600, threshold_planarity: float = 0.3, normal_threshold_deg: float = 5.0, depth_threshold: float = 0.025, neighbor_match_count_thresh: int = 8)

Options for the MoGe 4-head planarity detector. The segmentation defaults are the canonical parameters of the pxwplanar benchmark.

__init__(model_path: str = 'alpayozkan/pxwplanar-moge2-planarity', device: str = 'cuda', num_tokens: int = 1600, threshold_planarity: float = 0.3, normal_threshold_deg: float = 5.0, depth_threshold: float = 0.025, neighbor_match_count_thresh: int = 8) → None

depth_threshold: float = 0.025

device: str = 'cuda'

model_path: str = 'alpayozkan/pxwplanar-moge2-planarity'

neighbor_match_count_thresh: int = 8

normal_threshold_deg: float = 5.0

num_tokens: int = 1600

threshold_planarity: float = 0.3

Base interface

class limap.image.groups.planelib.base_plane_detector.BasePlaneDetector(options: BasePlaneDetectorOptions)

__init__(options: BasePlaneDetectorOptions)

detect_plane_mask(image_name: Path) → ndarray: Single-view plane segmentation. Filter and ensure continous labels.

detect_plane_mask_with_normals(image_name: Path) → tuple[ndarray, ndarray | None]

Detect planes and optionally return normal map.

Returns:: (H, W) filtered plane mask with continuous labels normal_map: (H, W, 3) normal map or None if not supported
Return type:: plane_mask

get_module_name() → str: Virtual method (need to be implemented) - return the name of the module

class limap.image.groups.planelib.base_plane_detector.BasePlaneDetectorOptions(min_num_pixels: int = 500, weight_path: pathlib.Path | None = None)

__init__(min_num_pixels: int = 500, weight_path: Path | None = None) → None

min_num_pixels: int = 500

weight_path: Path | None = None

Conversion utilities

limap.image.groups.planelib.convert_plane_mask_to_groups2d(plane_mask: ndarray, points: ndarray, lines: list[Line2d], min_line_overlap_length: float = 40.0, dilation_radius: float = 2.0, normal_map: ndarray | None = None) → list[Group2d]

Convert a plane segmentation mask to Group2d objects.

Parameters:

plane_mask – (H, W) int array with plane labels (0 = background)
points – (N, 2) array of 2D point coordinates
lines – list of Line2d objects
min_line_overlap_length – minimum overlap length to associate a line
dilation_radius – radius for dilating plane regions
normal_map – optional (H, W, 3) array of per-pixel normal vectors. If provided, average normals are computed per plane and stored as Group2d params.

Returns:

List of Group2d objects, one per plane label (1..num_planes)

Visualization

limap.image.groups.planelib.visualize_top_components(segmentation: ndarray, top_n: int = 10, ignore_label: int = 0, colormap: str = 'hls', background_image: ndarray | None = None, alpha: float = 0.5)

Visualize top-n largest plane segments with distinct colors.

Parameters:

segmentation – (H, W) segmentation map with plane IDs
top_n – Number of top planes to show
ignore_label – Label to treat as background (default 0)
colormap – Seaborn color palette name
background_image – Optional (H, W, 3) BGR image to blend with
alpha – Blending factor for overlay (0=background only, 1=colors only)

Returns:

(H, W, 3) BGR image

Return type:

colored_seg

limap.image.groups.planelib.visualize_single_component(segmentation: ndarray, target_label: int, color: tuple[int, ...] = (255, 0, 0), background_image: bool | None = None, alpha: float = 0.6): visualization of only one connected component. segmentation : 2D array of labels target_label : label id to highlight color : RGB tuple (default red)

limap.image.groups.planelib.visualize_plane_tracks(db_path: Path, structure_db_path: Path, workspace_path: Path, image_dir: Path, neighbors: dict, recon, output_dir: Path)