cfaed Seminar Series

Abstract 

6D pose estimation is the task of inferring the 3D position and 3D orientation of an object or the camera from an input image. It is an important problem in computer vision with applications in robotics, augmented reality and human-computer interaction. In this talk, we present an approach to object pose estimation which combines machine learning with geometry. A learnable regressor predicts dense correspondences (so called object coordinates) between an input image and the object surface. Because the object coordinate prediction contains outliers, we solve for the final pose using RANSAC. We show how this pipeline can be trained end-to-end using a new, differentiable RANSAC formulation, which we call DSAC.

Finally, we will discuss some possible future research directions.

Everybody is welcome!