Category-Level 6D Pose and Size Estimation using NOCS 

A PyTorch implementation of  Normalized Object Coordinate Space for Category-Level 6D Object Pose​ and Size Estimation, Wang et al. CVPR 2019​

Pytorch Implementation

GitHub

Original Work

Link

Abstract

We re-implented the main portion of the NOCS paper as a part of the DeepRob course at the UMN.  This was done in PyTorch and allows users to use weights from our training or train a model of their own. The results were comparable to the original work.

Video Presentation

G6_Final_Video.mp4

Results

 Evaluation

Here we show the accuracies of category poses as the threshold for rotation and translation varies. Additionally, the accuracy of 3D bounding box is plotted as the IoU with the ground truth is increased