which type of camera image should be feed to autoware?

edit

I was able to run KITTI dataset on Autoware, especially vision detect node. When I play the rosbag, I have to map the camera image topic to /image_raw to feed vision_yolo3_detect node. However, there's four types of camera image provided by KITTI: camera_gray_left, camera_gray_right, camera_color_left, camera_color_right. I'm not sure which type should I map to /image_raw and which impact has between right/left and color/gray?