The existing deep estimation networks often overlook the issue of computational efficiency while pursuing high accuracy. This paper proposes a lightweight self-supervised network that combines ...