Kevin's photo

Professor

Kai Xu (徐凯) is a Professor at the School of Computer Science, National University of Defense Technology (NUDT). He is also an Adjunct Professor of Simon Fraser University. He earned his Ph.D in computer science in 2011 from NUDT. From 2008 to 2010, He worked as a visiting PhD in the GrUVi lab of Simon Fraser University, supervised by Richard (Hao) Zhang. During 2012-2014, he conducted postdoctoral research at the VCC@SIAT, working with Baoquan Chen. During 2017-2018, he visited the Vision and Robotics Group at Princeton University, working with Thomas Funkhouser and Szymon Rusinkiewicz. Here is Kai's CV.

Kai is leading the iGRAPE Lab @ NUDT, which conducts research in the areas of computer graphics and computer vision. The current directions of interest include data-driven shape analysis and modeling, 3D vision and robot perception & navigation, etc. For more details, please refer to our publications.

gmail


       Representative research projects (click on the images to view the project pages):

mipsfusion_siga23irbpp_tog23ibsgrasp_sig22rosefusion_sig21

symmetrynet_sig20scores_siga18scores_siga18fglabel_siga18

projstyle_tog18nbo_sig18grass_sig17grass_sig17

scsc_sig17nbv_siga16pr2scene_siga15geotopo_siga15

focal_sig14topo_sig14symbr_sig13coh_sig13

msym_siga12civil_sig12photo_sig12style_siga10

pirs_siga09fast_siga09


Recent work

  • [New] Aug 1, 2024: Our LLM-enhanced household rearrangement paper is conditionally accepted to SIGGRAPH Asia 2024. Please check out more recent papers here.
  • Aug 5, 2023: Our generic shape bin packing paper is accepted by TOG and will be presented at SIGGRAPH 2024. Our MIPS-Fusion paper is conditionally accepted to SIGGRAPH Asia 2023. Three papers are accepted to ICCV 2023. Please check out details here.
  • Feb 28, 2023: Five papers are accepted to CVPR 2023, two to ICRA 2023 and one to AAAI 2023 as an oral paper. Please check out details here.
  • Aug 29, 2022: One paper on single-view symmetry detection is accepted to TPAMI and one about fast-camera-motion realtime RGB-D reconstruction is accepted to TRO. Please check out details here.
  • April 10, 2022: Five papers are accepted to CVPR 2022, one paper on DRL-based online bin packing accepted to ICLR 2022, and one paper on modeling grasping by Interaction Bisector Surface (IBS) accepted to SIGGRAPH 2022. Please check out details here.
  • May 10, 2021: Our paper on fast-camera-motion RGBD reconstruction is accepted to SIGGRAPH 2021. Please check out details here.
  • Feb. 24, 2021: Two papers accepted to CVPR 2021. Please check out details here.
  • Aug. 2, 2020: Our paper on end-to-end learnable symmetry prediction from single-view RGB-D images is accepted to SIGGRAPH Asia 2020. Please check out details here.
  • Jun. 15, 2020: Our paper on differentiable grasp planner is accepted to RSS 2020. Please check out details here.
  • Feb. 24, 2020: Five papers accepted to CVPR 2020. Please check out details here.
  • Dec. 20, 2019: We are pleased to annonce the release of the PartNet-Symh dataset (used by our CVPR 2019 paper PartNet). This dataset augments the PartNet dataset by adding a hierarchical organization of fine-grained parts for each shape. In general, PartNet-Symh can be used to train any model for encoding/decoding part-based structures based on Recursive Neural Networks, e.g., GRASS.
  • June. 18, 2019: I am glad to annonce the release of the source codes and benchmarks of our CVPR 2019 papers: Shape2Motion (code, data); PartNet (code and data) and HierContext (code). Thanks go to the students and collaborators. Shape2Motion and HierContext are also invited to be presented at the 3D Scene Generation Workshop at CVPR 2019.
  • May. 6, 2019: Daniel Ritchie, Sid Chaudhuri, Richard Zhang and I organized a tutorial at Eurographics 2019. Please check out details here.
  • Mar. 24, 2019: Our multi-robot scene reconstruction paper is conditionally accepted to SIGGRAPH 2019. Please check out details here.
  • Mar. 2, 2019: Three papers are accepted to CVPR 2019, including one oral (Shape2Motion) and two posters (HierContext and PartNet).
  • Oct. 17, 2018: Our two SIGGRAPH Asia 2018 papers, deep recursive 3D shape composition and fine-grainded component labeling for 3D shapes, have released the full source codes along with datasets.
  • Jun. 10, 2018: Our PlaneMatch paper is accepted as an oral paper to ECCV 2018. The paper is about 3D reconstruction of RGB-D sequences through learning a deep model for co-planarity-based matching. The source code is available here.
  • Jun. 10, 2018: Our GRASS paper is selected as a poster at the Bridges to 3D Workshop at CVPR 2018. See the poster.
  • Mar. 27, 2018: Our TOG paper, "Semi-Supervised Co-Analysis of 3D Shape Styles from Projected Lines", is awarded the Graphics Replicability Stamp! We have built a website of online 3D shape style analysis. You are welcomed to test our method by uploading your own 3D models.
  • Mar. 26, 2018: Three papers are to be presented at SIGGRAPH 2018. Other newly accepted papers include one CVPR 2018 paper, one ICRA 2018 paper and one AAAI 2018 paper. Please check out details here.
  • Aug. 3, 2017: Our paper "Autonomous Reconstruction of Unknown Indoor Scenes Guided by Time-varying Tensor Fields" is accepted to SIGGRAPH Asia 2017. Please see more details here. The source code of this work has been released as a package on ROS.
  • Apr. 1, 2017: Our GRASS (Generative Recursive Autoencoders for Shape Structures) paper has been selected as one of six SIGGRAPH 2017 papers featured for press release including EurekAlert, TechXplore and XinhuaNet, etc.
  • Jun. 20, 2017: The source code of our SIGGRAPH 2017 paper "GRASS: Generative Recursive Autoencoders for Shape Structures" is now available via github. UPDATE: A PyTorch implementation (currently with only the VAE part) is available (thanks to Chenyang Zhu's original implementation).
  • Apr. 23, 2017: Two papers are accepted to SIGGRAPH 2017. One is about learning generative models of 3D shape structures and the other is about structure-aware shape correspondence. Check out details here.
  • Oct. 20, 2016: The source code of 3D Recurrent Attention Model for Object Identification proposed in our SIGGRAPH Asia 2016 paper is made available via github. Please also check out a follow-up work on this topic here.
  • Aug. 18, 2016: Our course, "Data-Driven Shape Analysis and Processing", is accepted to SIGGRAPH Asia 2016. This course will be given together with Vladimir Kim, Qixing Huang, Evangelos Kalogerakis and Niloy Mitra.
  • Aug. 10, 2016: Our paper, "3D Attention-Driven Depth Acquisition for Object Identification", is conditionally accepted to SIGGRAPH Asia 2016, to be held in Macao this December. Check more details here.
  • Jun. 16, 2016: Two papers,"Mobility Fitting using 4D RANSAC" and "CustomCut: On-demand Extraction of Customized 3D Parts with 2D Sketches", are accepted by SGP 2016. See details here.
  • Oct. 30, 2015: Our survey paper, "Data-Driven Shape Analysis and Processing", has been accepted by Computer Graphics Forum. See more details here.
  • Oct. 30, 2015: The source code of object analysis (object-level segmentation and entropy-based NBP selection) of our robot-operated scene scanning paper is available for downloading.
  • Aug. 12, 2015: Two papers, "Autoscanning for Coupled Scene Reconstruction and Proactive Object Analysis" and "Deformation-Driven Topology-Varying 3D Shape Correspondence", are conditionally accepted to SIGGRAPH Asia 2015. Please see more details here.
  • Aug. 9, 2015: Our paper on Projective Feature Learning for 3D Shapes is accepted to Pacific Graphics 2015. Source code is available for downloading.
  • Aug. 7, 2015: The source code of mesh segmetation based on Extreme Learning Machine of our SGP 2014 paper is now available. The training time of our method is 100x faster than the state-of-the-art methods.
  • Feb. 25, 2015: Our survey paper "Data-Driven Shape Analysis and Processing" is now online via arXiv.org: arXiv:1502.06686. The wikipage is available online.
  • Jan. 5, 2015: One paper on Skeleton-Intrinsic Symmetrization is conditionally accepted to Eurographics 2015.
  • May. 17, 2014: The source code of focal analysis of our SIGGRAPH 2014 paper is available.
  • Mar. 28, 2014: Two papers, "Organizing Heterogeneous Scene Collections through Contextual Focal Points" and "Topology-Varying 3D Shape Creation via Structural Blending", are conditionally accepted to SIGGRAPH 2014. See more details here.
  • Jun. 27, 2013: The source code and data of our facade structure analysis paper is now available.
  • Jun. 14, 2013: Our paper on outdoor scene consolidation is accepted to SGP 2013. Another paper on sketch-based 3D object design is accepted to CGF (with minor revision).
  • Apr. 22, 2013: The project pages of our SIGGRAPH 2013 papers are available; see symmetry maximization and conshier.
  • Apr. 6, 2013: Two papers are conditionally accepted to SIGGRAPH 2013: 1) Layered Analysis of Irregular Facades via Symmetry Maximization; 2) Co-Hierarchical Analysis of Shape Structures. Please see more details here.
  • Sep. 9, 2012: The datasets in our set evolution paper are available for downloading.
  • Aug. 23, 2012: Our paper "Multi-Scale Partial Intrinsic Symmetry Detection" is conditionally accepted to SIGGRAPH Asia 2012.
  • Jun. 8, 2012: Project page of our SIGGRAPH 2012 paper is on.
  • Apr. 3, 2012: Our paper "Fit and Diverse: Set Evolution for Inspiring 3D Shape Galleries" is conditionally accepted to SIGGRAPH 2012.
  • Apr. 30, 2011:  Project page of our SIGGRAPH 2011 paper is now on.
  • Mar. 31, 2011:  Our paper "Photo-Inspired Model- Driven 3D Object Modeling" is conditionally accepted to SIGGRAPH 2011.
  • Jan. 31, 2011:  The project page of symmetry hierarchy is now available from Eric's website.
  • Dec. 1, 2010:  Our paper "Symmetry Hierarchy of Man-Made Objects" is conditionally accepted to Eurographics 2011.
  • Aug. 28, 2010:  Project page for "Style-Content Separation by Anisotropic Part Scales" is on.
  • Aug. 6, 2010:  Our paper on Style-Content Separation is conditionally accepted to SIGGRAPH Asia 2010.
  • Oct. 23, 2009:  The source code of curve orientation in our feature-aligned texturing paper is now available.

Recent activities