tweaking the Semantic Segmentation

[samhumeau]

Dirty hack the the semantic segmentation camera.

You can use this wrapper to interpret the result:

class ColorModifierWrapper(object):
    """ We can not instanciate a Carla image from the python side, so this is
        an helper to convert what's out of the semantic segmentation now
    """

    def __init__(self, carla_image):
        self.carla_image = carla_image
        self.original_colors = [ColorModifierWrapper.clone_color(carla_image[i])
                                for i in range(len(carla_image))]
        self.color_map = {r:carla.Color(r = (r*13+16) % 256,
                                        g = (142+r*21) % 256,
                                        b = (89+r*47) % 256)
                          for r in range(256)}
        self.blank = carla.Color(r=255,b=255,g=255)
        self.black = carla.Color(r=0,b=0,g=0)

    @staticmethod
    def clone_color(color):
        return carla.Color(r=color.r, g=color.g, b=color.b, a=color.a)

    def view_as_semantic_seg(self):
        for i, color in enumerate(self.original_colors):
            new_color = ColorModifierWrapper.clone_color(color)
            if color.r >=16 and color.r < 128:
                new_color.r = 4 # pedestrian
            if color.r >=128 and color.r < 240:
                new_color.r = 10 # vehicle
            self.carla_image[long(i)] = new_color
        self.carla_image.convert(carla.ColorConverter.CityScapesPalette)
        return self.carla_image


    def view_as_instance_seg(self):
        for i, color in enumerate(self.original_colors):
            if color.r < 16 and color.r != 4:
                self.carla_image[long(i)] = self.blank # Not a pedestrian nor a car
            elif color.r == 4:
                self.carla_image[long(i)] = self.black # rider
            else:
                self.carla_image[long(i)] = self.color_map[color.r]
        return self.carla_image