Jiaxin Lu, Hao Kang, Haoxiang Li, Bo Liu, Yiding Yang, Qixing Huang, Gang Hua
eccv 2024

We introduce a unified diffusion-based dexterous grasp generation model, UGG. Our all-transformer architecture unifies the information from the object, the hand, and the contacts. A proposed lightweight discriminator, benifiting from simulated data, pushes for a high success rate while preserving high diversity. Our model achieves state-of-the-art dexterous grasping on the large-scale DexGraspNet dataset while facilitating human-centric object design, marking a significant advancement in dexterous grasping research.

Jiaxin Lu*, Yifan Sun*, Qixing Huang
NeurIPS 2023

We show how to jointly learn segmentation and matching and seamlessly integrate feature matching and rigidity constraints for the learning of multiple fracture assembly. We evaluate Jigsaw on the Breaking Bad dataset and achieve superior performance compared to state-of-the-art methods. Our method also generalizes well to diverse fracture modes, objects, and unseen instances. To the best of our knowledge, this is the first learning-based method designed specifically for 3D fracture assembly over multiple pieces.

Zetian Jiang*, Jiaxin Lu*, Tianzhe Wang, Junchi Yan
Submitted to T-PAMI, Under Review

We analyze the partial matching problem and reveal the limitations of existing methods on distinguishing unmatched inlier and outliers. Based on a universe matching perspective, we build an end-to-end learning pipeline including universe metric learning scheme and outlier-aware loss. Our method UPM significantly outperforms SOTA on main-stream datasets. It is also the first deep learning method that can deal with different complex extension cases simultaneously, reaching a notably accelerated matching process and less space utilization as well.

Jiaxin Lu*, Zetian Jiang*, Tianzhe Wang, Junchi Yan
ICLR 2024

We propose an efficient MM-based algorithm M3C that iteratively tackling graph matching and clustering problem. Based on M3C, an unsupervised learning model UM3C is further developed which is equipped with our devised edge-wise affinity learning and pseudo label selection techniques. Experimental results on public benchmarks show that our method notably surpasses the state-of-the-art methods in both accuracy and efficiency, where our unsupervised model even exceeds supervised methods.