Why is +0.5 and dilate with 0.5 a better solution in mask_to_polygons function in detectron2.utils.visualizer?

[esmeecerfonteijn]

The mask_to_polygons function that is defined in the detectron2.utils.visualizer source code (see https://detectron2.readthedocs.io/en/latest/_modules/detectron2/utils/visualizer.html) contains a comment that says that first adding +0.5 to the coordinates and then dilating the polygon with a factor 0.5, is a better solution than only adding +0.5 to the coordinates from the OpenCV contour.

• Why would that a better solution?

• What is the reason for the current implementation instead of the proposed better solution?

Current implementation: res = [x + 0.5 for x in res if len(x) >= 6]
Adding +0.5 to the OpenCV coordinates translates the coordinates to the pixel center, that is at (0.5, 0.5) of each pixel.