Learning-AI

MVRA: Multi-View Reprojection Architecture for Orientation Estimation

November 2019

tl;dr: Build the 2D/3D constraints optimization into neural network and use iterative method to refine cropped cases.

Overall impression

This paper is heavily based on deep3Dbox and adds a few improvement to handle corner cases.

The paper has a very good introduction to mono 3DOD methods.

Key ideas

Technical details

Notes