About the paper.

[TongWeiDP]

Greater works! There is something I don’t understand in your paper. I want to know if you combine semantic segmentation and depth estimation together for training? The cityscapes data seems to have only semantic segmentation with no depth information. The images of cityscapes are also go through the branch of the depth prediction network？ In addition, which of kitti's semantic segmentation datasets did you use specifically? I find that the pixel-level data set of kitti is only 342MB. Did you use this data set? Looking forward to your reply.Thanks.