OnePose++: Keypoint-Free One-Shot Object Pose Estimation without CAD Models
Nov 28, 2022
Speakers
Jiaming Sun
Speaker · 0 followers
Yuang Wang
Speaker · 0 followers
Xiaowei Zhou
Speaker · 0 followers
About
We propose a new method for object pose estimation without CAD models. The previous feature-matching-based method OnePose has shown promising results under a one-shot setting, eliminating the object-specific training and aiming for the pose estimation of arbitrary unseen objects. However, OnePose relies on detecting repeatable image keypoints and is thus prone to fail on low-textured objects. We propose to remove the need for keypoint detection with a keypoint-free SfM and pose estimation pipeline. Built upon the keypoint-free feature matching method LoFTR, our novel keypoint-free SfM reconstructs object semi-dense point clouds, which provides complete 3D models for pose estimation of low-textured objects. During pose estimation, our novel keypoint-free 2D-3D matching network matches the reconstructed point cloud with the query image to build 2D-3D correspondences for object pose estimation. The keypoint-free SfM and pose estimation pipeline enables pose estimation of low-textured objects under the one-shot setting, making it more applicable in the real world. We demonstrate that the proposed method outperforms existing one-shot CAD-model-free methods by a large margin and even achieves comparable results with CAD-model-based methods on LINEMOD. We also collect a new dataset composed of 80 sequences of 40 low-textured objects to boost future research on one-shot object pose estimation. Code and data will be made available upon the publication of this paper.We propose a new method for object pose estimation without CAD models. The previous feature-matching-based method OnePose has shown promising results under a one-shot setting, eliminating the object-specific training and aiming for the pose estimation of arbitrary unseen objects. However, OnePose relies on detecting repeatable image keypoints and is thus prone to fail on low-textured objects. We propose to remove the need for keypoint detection with a keypoint-free SfM and pose estimation pipeli…
Organizer
NeurIPS 2022
Account · 952 followers
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Flamingo: a Visual Language Model for Few-Shot Learning
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Receding Horizon Inverse Reinforcement Learning
Watch later
Yiqing Xu, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Revisiting Graph Contrastive Learning from the Perspective of Graph Spectrum
Watch later
Nian Liu, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Subgame Solving in Adversarial Team Games
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Structure Inducing Pre-training
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Positively Weighted Kernel Quadrature via Subsampling
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%