3Dwe at CVPR 2024
This year, Jun 17th through Fri Jun 21st the Seattle Convention Center becomes the global stage for advancements in computer vision at the CVPR 2024 (Computer Vision and Pattern Recognition) conference. 3Dwe takes part with contributions in three papers pushing boundaries of 3D computer vision.
HouseCat6D: A Large-Scale Multi-Modal Category Level 6D Object Perception Dataset with Household Objects in Realistic Scenarios
Looking for data to evaluate your pose estimation or object grasping pipeline? Check out HouseCat6D, a CVPR2024 highlight featuring 160K poses, 10M grasps, 41 scenes, 194 objects, 10 household categories, and 3 modalities. This dataset includes synchronized RGB, depth from active stereo, and polarimetric RGB+P images in realistic scenarios with diverse objects and complex photometric challenges.
MatchU: Matching Unseen Objects for 6D Pose Estimation from RGB-D Images
Discover faster and more accurate pose estimation of unseen objects with MatchU. This method uses a fuse-describe-match framework that combines geometry and image features for 6D pose estimation from RGB-D images without the need for resource-intensive training. MatchU's novel attention-based mechanism fuses cross-modal data, enabling superior generalizability and performance.
SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation
Improve your category-level pose pipeline with SecondPose. SecondPose integrates geometric (PPF) and semantic (DINOv2) features for SE(3)-consistent dual-stream feature fusion, enhancing 6D pose estimation. Extensive experiments show SecondPose's superior performance, achieving significant advancements on datasets like NOCS-REAL275 and HouseCat6D.