This might be due to quality of the dataset because Nvidia seems to be not using copyrighted commercial recordings (if I read their paper properly). It is difficult to compete with those who have used larger and higher quality dataset without permission.