Why does the semantic sensor output perspective differ from RGB and depth sensors?

[BaiLing09]

Habitat-Lab and Habitat-Sim versions
Habitat-Lab: tags challenge-2022

Habitat-Sim: tags challenge-2022

Datasets: MP3D sence and its object navigation task episodes

❓ Questions and Help
Questions: the semantic sensor output perspective differ from RGB and depth sensors. (I try to print the sensor state. This three kinds of sensor are same states)

Following is my code:
config_paths = "/Configs/Object_Nav_mp3D.yaml"
config = get_config(config_paths)

hab_env = Env(config=config)
print(hab_env._sim.get_agent_state())

action = {1: 'MOVE_FORWARD', 2: 'TURN_LEFT', 3: 'TURN_RIGHT'}
ep_i = 0
while ep_i < 10:
observations = hab_env.reset()
start = 0
while start < 15:
obs = hab_env.step(action[random.choice([1, 2, 3])])
sematic, depth = dis(obs['semantic'], obs['depth']) # dis is a function only for creating Visual image
cv.imshow("RGB", obs['rgb'])
cv.imshow("Semantic", sematic)
cv.imshow("depth", depth)
cv.waitKey(500)
start += 1

Following is my config yaml:

Following is the running result:

[JahaoZhang]

How did you solve this problem? I'm facing the same issue.