trainning batch size

[vehxianfish]

Thank you for your wonderful work. The paper you published is trained with batch_size16. But when I train on a single 2080Ti with batch_size =2, the performance is better than on four cards. So your batch_size =16 is trained on a single card or on multi-cards. And I submit the result to the Kitti test server, the moderate level is only 80.04 . It confused me for about two months. So can you give a solution? Thank you very much.
Best regards!