Convert Between Polygons and Masks

Polygons and RLE masks are two common representations for regions. Here’s how to convert between them.

Polygon to mask

import numpy as np
from rlemasklib import RLEMask

# Polygon as Nx2 array of (x, y) vertices
polygon = np.array([[10, 10], [100, 10], [100, 80], [10, 80]])

mask = RLEMask.from_polygon(polygon, imshape=(480, 640))

Multiple polygons (with holes)

For a polygon with holes, pass a list of vertex arrays. The first is the outer boundary, the rest are holes:

outer = np.array([[0, 0], [100, 0], [100, 100], [0, 100]])
hole = np.array([[30, 30], [70, 30], [70, 70], [30, 70]])

mask = RLEMask.from_polygon([outer, hole], imshape=(200, 200))

Union of separate polygons

For multiple disconnected regions, create masks and combine:

polygons = [poly1, poly2, poly3]
masks = [RLEMask.from_polygon(p, imshape) for p in polygons]
combined = RLEMask.union(masks)

From shapely geometry

import shapely

# Shapely Polygon
shape = shapely.Polygon([(10, 10), (100, 10), (100, 80), (10, 80)])
mask = RLEMask.from_polygon(np.array(shape.exterior.coords), imshape)

# MultiPolygon
masks = [RLEMask.from_polygon(np.array(p.exterior.coords), imshape)
         for p in multi_polygon.geoms]
combined = RLEMask.union(masks)

COCO format

COCO stores polygons as flat lists [x1, y1, x2, y2, ...]:

# From COCO polygon
coco_poly = [10, 10, 100, 10, 100, 80, 10, 80]
polygon = np.array(coco_poly).reshape(-1, 2)
mask = RLEMask.from_polygon(polygon, imshape)