Diff9D: Diffusion-Based Domain-Generalized Category-Level 9DoF Object Pose Estimation (Code)
Jian Liu, Wei Sun, Hui Yang, Pengchao Deng, Chongpei Liu, Nicu Sebe, Hossein Rahmani, Ajmal Mian

• We propose an effective diffusion model to redefine 9DoF object pose estimation from a generative perspective. Diff9D is a simple yet effective prior-free domain-generalized (sim2real) category-level 9DoF object pose generator. By employing the denoising diffusion implicit model, we demonstrate that the reverse diffusion process can be executed in as few as 3 steps, achieving near real-time performance.

Novel Object 6D Pose Estimation with a Single Reference View (Code)
Jian Liu, Wei Sun, Kai Zeng, Jin Zheng, Hui Yang, Lin Wang, Hossein Rahmani, Ajmal Mian

• We propose a single reference view-based CAD model-free novel object 6D pose estimation method. SinRef-6D is simple yet effective and can simultaneously eliminate the need for object CAD models, dense reference views, and model retraining, offering enhanced efficiency and scalability while demonstrating strong generalization to potential real-world robotic applications.

Deep Learning-Based Object Pose Estimation: A Comprehensive Survey
Project Page
Jian Liu, Wei Sun, Hui Yang, Zhiwen Zeng, Chongpei Liu, Jin Zheng, Xingyu Liu, Hossein Rahmani, Nicu Sebe, Ajmal Mian

• We present a comprehensive survey of deep learning-based object pose estimation methods. This survey covers all three problem formulations in the domain, including instance-level, category-level, and unseen object pose estimation. We hope to provide readers with a complete picture of the research progress of deep learning-based object pose estimation.

MH6D: Multi-Hypothesis Consistency Learning for Category-Level 6D Object Pose Estimation (Code)
Jian Liu, Wei Sun, Chongpei Liu, Hui Yang, Xing Zhang, Ajmal Mian

• We propose a multi-hypothesis consistency learning framework for category-level 6D object pose estimation, which utilizes a parallel consistency learning structure, alleviating the uncertainty problem of single-shot feature extraction and promoting self-adaptation of domain to reduce the synthetic-to-real domain gap.

Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category-Level 6D Pose Estimation (Code)
Jian Liu, Wei Sun, Chongpei Liu, Xing Zhang, Qiang Fu

• A transformer-guided shape reconstruction network is proposed to reconstruct the NOCS shape of intra-class known objects, which can fully use the prior feature, current observation feature, and their feature difference by internal self-attention, as well as strengthen their correlation by mutual cross-attention. By doing so, the shape variation can be explicitly highlighted.

HFF6D: Hierarchical Feature Fusion Network for Robust 6D Object Pose Tracking
Jian Liu, Wei Sun, Chongpei Liu, Xing Zhang, Shimeng Fan, Wei Wu

• We propose a lightweight and robust hierarchical feature fusion network for 6D object pose tracking. It establishes sufficient spatial-temporal information interaction between adjacent frames and explicitly highlights the feature differences between adjacent frames, thus improving the robustness of relative pose estimation in challenging scenes.

Robotic Continuous Grasping System (Demo can be seen through link1 or link2)

• We build an end-to-end robotic continuous grasping system, which achieves high-accuracy 6D pose estimation for multiple intra-class unknown objects and high-efficiency robotic grasping in 3D space. For continuous grasping, we propose a low-computation and effective grasping strategy based on the pre-defined vector orientation, and develop a GUI for monitoring and control.

Robotic Continuous Picking System (Demo can be seen through link)

• We develop an object pose-guided robotic picking system comprising both hardware and software components. The hardware for the robotic picking system is composed of an Intel RealSense L515 RGB-D camera, a Yaskawa robot MOTOMAN-MH12, an electric parallel gripper DH-PGI140-80, and a host computer. For software, we develop a GUI comprising three parts: calibration module, server-client TCP communication module, and robotic picking module.