Matplotlib创建语义分割用Cityscapes_Palette_Map离散colormap

以下是Python生态下两种常用的实现方案，完全适配0-22类的离散标签到固定颜色的一对一映射需求，不会出现连续cmap的插值混色问题：

方案1：Matplotlib 自定义离散ListedColormap #

适用于需要用Matplotlib绘图渲染语义分割结果的场景，ListedColormap是Matplotlib专门提供的离散类别映射接口：

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap, BoundaryNorm

# 按0-22类别顺序排列的Carla语义标签RGB颜色（0-255格式，和Cityscapes标准配色对齐）
carla_semantic_colors = [
    (0, 0, 0),        # 0: Unlabeled
    (70, 70, 70),     # 1: Building
    (100, 40, 40),    # 2: Fence
    (55, 90, 80),     # 3: Other
    (220, 20, 60),    # 4: Pedestrian
    (153, 153, 153),  # 5: Pole
    (157, 234, 50),   # 6: Road line
    (128, 64, 128),   # 7: Road
    (244, 35, 232),   # 8: Sidewalk
    (107, 142, 35),   # 9: Vegetation
    (0, 0, 142),      # 10: Car
    (102, 102, 156),  # 11: Wall
    (220, 220, 0),    # 12: Traffic sign
    (70, 130, 180),   # 13: Sky
    (81, 0, 81),      # 14: Ground
    (150, 100, 100),  # 15: Bridge
    (230, 150, 140),  # 16: Rail track
    (180, 165, 180),  # 17: Guard rail
    (250, 170, 30),   # 18: Traffic light
    (110, 190, 160),  # 19: Static
    (170, 120, 50),   # 20: Dynamic
    (45, 60, 150),    # 21: Water
    (145, 170, 100)   # 22: Terrain
]
# 转归一化到[0,1]的格式适配Matplotlib要求
carla_colors_norm = np.array(carla_semantic_colors) / 255.0
# 生成离散cmap
cityscapes_cmap = ListedColormap(carla_colors_norm, name='carla_cityscapes')
# 定义边界，保证每个整数标签严格对应对应颜色，避免插值偏移
norm = BoundaryNorm(boundaries=np.arange(-0.5, 23, 1), ncolors=cityscapes_cmap.N)

# 测试用例
semantic_mask = np.random.randint(0, 23, size=(200, 200))
plt.imshow(semantic_mask, cmap=cityscapes_cmap, norm=norm)
plt.axis('off')
plt.show()

方案2：Numpy索引直接映射（全场景适配） #

不需要依赖Matplotlib，直接生成可用于OpenCV、PIL等库的RGB语义掩码，运算效率更高：

import numpy as np
import cv2

# 按0-22类别顺序排列的RGB颜色（0-255格式）
carla_semantic_colors = np.array([
    (0, 0, 0),        # 0: Unlabeled
    (70, 70, 70),     # 1: Building
    (100, 40, 40),    # 2: Fence
    (55, 90, 80),     # 3: Other
    (220, 20, 60),    # 4: Pedestrian
    (153, 153, 153),  # 5: Pole
    (157, 234, 50),   # 6: Road line
    (128, 64, 128),   # 7: Road
    (244, 35, 232),   # 8: Sidewalk
    (107, 142, 35),   # 9: Vegetation
    (0, 0, 142),      # 10: Car
    (102, 102, 156),  # 11: Wall
    (220, 220, 0),    # 12: Traffic sign
    (70, 130, 180),   # 13: Sky
    (81, 0, 81),      # 14: Ground
    (150, 100, 100),  # 15: Bridge
    (230, 150, 140),  # 16: Rail track
    (180, 165, 180),  # 17: Guard rail
    (250, 170, 30),   # 18: Traffic light
    (110, 190, 160),  # 19: Static
    (170, 120, 50),   # 20: Dynamic
    (45, 60, 150),    # 21: Water
    (145, 170, 100)   # 22: Terrain
], dtype=np.uint8)

# 输入语义mask，形状为[H, W]，值为0-22的整数
semantic_mask = np.random.randint(0, 23, size=(200, 200), dtype=np.uint8)
# 直接索引映射得到RGB图像，形状为[H, W, 3]
rgb_mask = carla_semantic_colors[semantic_mask]

# OpenCV展示示例（需要把RGB转BGR）
cv2.imshow('semantic mask', cv2.cvtColor(rgb_mask, cv2.COLOR_RGB2BGR))
cv2.waitKey(0)

内容的提问来源于stack exchange，提问作者paul