6D Object Pose Estimation in the Wild. Self-supervised Geometric Correspondence for Category-level 6D Object Pose Estimation in the Wild
Kaifeng Zhang1, Yang Fu2, Shubhankar Borse3, Hong Cai3, Fatih Porikli3, Xiaolong Wang2
1Tsinghua University,2UC San Diego,3Qualcomm AI Research
This project determines the six degrees of freedom of an object in 3D space and can be used for a variety of tasks, such as object recognition, tracking, and manipulation. The challenge with 6D pose estimation is that it is often difficult to obtain accurate annotations, especially for in-the-wild images. This self-supervised learning approach is designed to overcome such challenges by learning from large-scale real-world object videos.
Key features:
- Learns from large-scale real-world object videos without the need for human annotations.
- Uses geometrical cycle-consistency losses to establish correspondences between 2D and 3D spaces
- Achieved SOTA performance.
Developed by Kaifeng Zhang, Yang Fu, Shubhankar Borse, Hong Cai, Fatih Porikli, Xiaolong Wang – Tsinghua University, UC San Diego, Qualcomm AI Research
Paper Project SEE Ai Generated Images
Try Q Blocks Today
Access on-demand GPU instances at upto 80% lower costs