Mono3D: Monocular 3D Object Detection for Autonomous Driving

August 2019

tl;dr: The pioneering paper on monocular 3dod, with tons of hand crafted feature.

Overall impression

The paper is quite innovative at the time being, but looks rather archaic three years later in 2019. In particular, it chose to use Fast (not Faster) RCNN as base net, together with tons of hand crafted features. Many of the cited work are pre-DL era. From industry’s standpoint of view, this paper is of little practical use as of 2019.

Key ideas

Technical details