Open headupinclouds opened 7 years ago
Need to benchmark EOS methods to see what can run feasibly on mobile devices. TODO: investigate better approaches to retrofit 3d eye models to mesh from 2d landmarks.
Use a model based approach to estimate ground truth from landmarks and then solve for pose as part of the cascaded pose regression. Various representations are possible for 3 DOF rotation:
http://imd.naist.jp/imdweb/pub/harada_fcv10/paper.pdf
We have shown experimental results of comparison of pose representations for global appearance-based pose estimation of an object.