Library of handy polygon related functions to speed up machine learning projects.
It was born as a replacement for cv2.fillPoly when generating masks for instance segmentation, without having to bring in all of opencv.
- draw_polygon
- find_contours
- polygon_area
- point_in_polygon
This library expects all polygons to be model as a list of paths, each path is a list of alternating x and y coordinates ([x1,y1,x2,y2,...]).
A simple triangle would be declared as:
triangle = [[50,50, 100,0, 0,0]]Complex polygons (holes and/or disjoints) follow the even-odd rule.
draw_polygon(mask: array[:, :], paths: path[]) -> array[:, :]
from upolygon import draw_polygon
import numpy as np
mask = np.zeros((100,100), dtype=np.int32)
draw_polygon(mask, [[50,50, 100,0, 0,0]], 1)Equivalent of calling cv2.fillPoly(mask, [np.array([[50,50], [100,0], [0,0]])], 1) or cv2.drawContours(mask, [np.array([[50,50], [100,0], [0,0]])], -1, 1, cv2.FILLED) when using opencv.
uPolygon is ~ 6 times faster than opencv for large random polygons with many intersecting lines. For smaller polygons or few intersections, uPolygon is half as fast as opencv.
find_contours(mask: array[:, :]) -> (array[:, :], path[:], path[:])
0 is treated as background, 1 is treated as foreground.
from upolygon import find_contours
import numpy as np
mask = np.array([
[0, 0, 0, 0, 0],
[0, 1, 1, 1, 0],
[0, 1, 1, 1, 0],
[0, 1, 1, 1, 0]
], dtype=np.uint8)
_labels, external_paths, internal_paths = find_contours(mask)Similar to OpenCV's cv2.findContours but lacking hierarchies. Also similar to BoofCV's LinearContourLabelChang2004 which is based on the same algorithm.
Note that currently the input mask to find_contour needs to be uint8.
rle_encode(mask: array[:,:]) -> list
Takes a 2-dim binary mask and generates a run length encoding according to the coco specs
~ 15 times faster than written in plain python