Haiping Wang
Research Assistant Professor @ HKUST (2026-Now)
I am an incoming Research Assistant Professor at ISD of The Hong Kong University of Science and Technology, closely working with Assistant Prof. Yuan Liu and Prof. Sai-Kit Yeung. Before that, I received my Ph.D. degree from LIESMARS of Wuhan University, supervised by Prof. Zhen Dong and Prof. Bisheng Yang. My research focuses on advanced 3D vision algorithms, including 3D reconstruction, 3D understanding, and Spatial Agent Systems.
News
- 2026/06: LEGO, 3D Open-vocabulary Understanding, is accepted by ECCV2026.
- 2026/02: SpatialLLM for zero-shot urban intelligence is accepted by JAG.
- 2026/01: SCoT, million-scale CoT dataset for 3D-LLM, is accepted by ICLR2026.
- 2026/01: RegScorer, learning to score transformations, is accepted by ISPRS J.
- 2025/11: GAGS, 3D Open-vocabulary Understanding, is accepted by AAAI2026.
- 2025/10: Neural City for urban spatial intelligence is accepted by The Innovation.
- 2025/06: VistaDream, training-free 3D scene generation, is accepted by ICCV2025.
- 2025/01: CityAnchor, 3D-LLM for city-scale grounding, is accepted by ICLR2025.
Honors & Awards
- Fundamental Project for PhD student from NSFC (¥300K) 2024
- Outstanding Poster of ChinaSI (Top 8) 2025
- National Scholarship for Ph.D. (Top 3%, Rank 1st in LISMARS) 2023
- Gold Medal of Smart City Technology Innovation Award (Rank 1st) 2023
- Outstanding Student Scholarship (Top 1%) 2020
- Outstanding Graduate of Wuhan University (Top 5%) 2020 & 2026
- The Best Paper Award at the National Lidar Conference (Top 10) 2018
- National Scholarship for Postgraduates (Top 3%) 2017 & 2018
Selected Publications (*Co-First, †Corresponding) (view all )
LEGO':' Leveled Language Gaussian Splatting
Yuning Peng, Haiping Wang†, Yuan Liu, Yipeng Lu, Zhen Dong, Bisheng Yang
European Conference on Computer Vision (ECCV) 2026
LEGO self-adaptively re-grades volatile multi-view SAM granularities into a unified, open-vocabulary, 3D-consistent hierarchy, such as the "flowerpot - bouquet - bud - petal" lineage.
LEGO':' Leveled Language Gaussian Splatting
Yuning Peng, Haiping Wang†, Yuan Liu, Yipeng Lu, Zhen Dong, Bisheng Yang
European Conference on Computer Vision (ECCV) 2026
LEGO self-adaptively re-grades volatile multi-view SAM granularities into a unified, open-vocabulary, 3D-consistent hierarchy, such as the "flowerpot - bouquet - bud - petal" lineage.
SpatialLLM':' From Multi-modality Data to Urban Spatial Intelligence
Jiabin Chen*, Haiping Wang*, Jinpeng Li, Yuan Liu†, Zhen Dong†, Bisheng Yang
International Journal of Applied Earth Observation and Geoinformation (IF:8.6) 2026
Structured descriptions of raw spatial data equip LLM with zero-shot execution of advanced spatial intelligence tasks, including urban planning, ecological analysis, traffic management, etc.. Multi-field knowledge, context length, and reasoning ability are key factors influencing LLM performances in urban analysis.
SpatialLLM':' From Multi-modality Data to Urban Spatial Intelligence
Jiabin Chen*, Haiping Wang*, Jinpeng Li, Yuan Liu†, Zhen Dong†, Bisheng Yang
International Journal of Applied Earth Observation and Geoinformation (IF:8.6) 2026
Structured descriptions of raw spatial data equip LLM with zero-shot execution of advanced spatial intelligence tasks, including urban planning, ecological analysis, traffic management, etc.. Multi-field knowledge, context length, and reasoning ability are key factors influencing LLM performances in urban analysis.
SCoT':' Teaching 3D-LLMs to Think Spatially with Million-scale CoT Annotations
Jinpeng Li*, Haiping Wang*, Jiabin Chen, Yuan Liu, Zhen Dong†, Bisheng Yang
International Conference on Learning Representations (ICLR) 2026
We present a million-scale 3D visual-language dataset with CoT annotations that unifies perception, analysis, and planning tasks to advance interpretable 3D intelligence..
SCoT':' Teaching 3D-LLMs to Think Spatially with Million-scale CoT Annotations
Jinpeng Li*, Haiping Wang*, Jiabin Chen, Yuan Liu, Zhen Dong†, Bisheng Yang
International Conference on Learning Representations (ICLR) 2026
We present a million-scale 3D visual-language dataset with CoT annotations that unifies perception, analysis, and planning tasks to advance interpretable 3D intelligence..
RegScorer ":" Learning to Select the Best Transformation of Point Cloud Registration
Xiaochen Yang*, Haiping Wang*, Yuan Liu, Bisheng Yang†, Zhen Dong
ISPRS Journal of Photogrammetry and Remote Sensing (ISPRS J, IF:12.2) 2026
RegScorer learns to score the candidate transformations for more reliable selection in pairwise point cloud registration.
RegScorer ":" Learning to Select the Best Transformation of Point Cloud Registration
Xiaochen Yang*, Haiping Wang*, Yuan Liu, Bisheng Yang†, Zhen Dong
ISPRS Journal of Photogrammetry and Remote Sensing (ISPRS J, IF:12.2) 2026
RegScorer learns to score the candidate transformations for more reliable selection in pairwise point cloud registration.
GAGS':' Granularity-Aware 3D Feature Distillation for Gaussian Splatting
Yuning Pang*, Haiping Wang*, Yuan Liu†, Chenglu Wen, Zhen Dong†, Bisheng Yang
AAAI Conference on Artificial Intelligence (AAAI) 2026
GAGS learns a 3D Gaussian field associated with semantic features, which enables accurate open-vocabulary 3D visual grounding in the scene.
GAGS':' Granularity-Aware 3D Feature Distillation for Gaussian Splatting
Yuning Pang*, Haiping Wang*, Yuan Liu†, Chenglu Wen, Zhen Dong†, Bisheng Yang
AAAI Conference on Artificial Intelligence (AAAI) 2026
GAGS learns a 3D Gaussian field associated with semantic features, which enables accurate open-vocabulary 3D visual grounding in the scene.
The Neural City':' A Next-generation Spatio-Temporal Intelligence Paradigm for Urban Holistic Governance
Zhen Dong*, Haiping Wang*, Zhe Chen, Chen Long, Yuning Peng, Yuan Liu, Fuxun Liang, Jian Zhou, Yiping Chen, Fan Zhang, Bisheng Yang†, Deren Li
The Innovation (IF: 25.7) 2025
We outline Neural City to enable end-to-end urban governance, seamlessly linking raw urban data to holistic urban governance, achieving "6W+4R" governance.
The Neural City':' A Next-generation Spatio-Temporal Intelligence Paradigm for Urban Holistic Governance
Zhen Dong*, Haiping Wang*, Zhe Chen, Chen Long, Yuning Peng, Yuan Liu, Fuxun Liang, Jian Zhou, Yiping Chen, Fan Zhang, Bisheng Yang†, Deren Li
The Innovation (IF: 25.7) 2025
We outline Neural City to enable end-to-end urban governance, seamlessly linking raw urban data to holistic urban governance, achieving "6W+4R" governance.
VistaDream':' Sampling multiview consistent images for single-view scene reconstruction
Haiping Wang, Yuan Liu†, Ziwei Liu, Zhen Dong†, Wenping Wang, Bisheng Yang
International Conference on Computer Vision (ICCV) 2025
VistaDream is a training-free framework to reconstruct a high-quality 3D scene from a single-view image. The key idea is to sample multi-view consistent high-quality images from pre-trained single-view diffusion models.
VistaDream':' Sampling multiview consistent images for single-view scene reconstruction
Haiping Wang, Yuan Liu†, Ziwei Liu, Zhen Dong†, Wenping Wang, Bisheng Yang
International Conference on Computer Vision (ICCV) 2025
VistaDream is a training-free framework to reconstruct a high-quality 3D scene from a single-view image. The key idea is to sample multi-view consistent high-quality images from pre-trained single-view diffusion models.
CityAnchor':' City-scale 3D Visual Grounding with Multi-modality LLMs
Jinpeng Li*, Haiping Wang*, Jiabin Chen, Yuan Liu†, Zhiyang Dou, Yuexin Ma, Sibei Yang, Yuan Li, Wenping Wang, Zhen Dong, Bisheng Yang†
International Conference on Learning Representations (ICLR) 2025
We present a two-stage (coarse-to-fine) 3D visual grounding system by tuning Large Vision Language Model (LVLM) to accurately find targets in city-scale point clouds from text descriptions.
CityAnchor':' City-scale 3D Visual Grounding with Multi-modality LLMs
Jinpeng Li*, Haiping Wang*, Jiabin Chen, Yuan Liu†, Zhiyang Dou, Yuexin Ma, Sibei Yang, Yuan Li, Wenping Wang, Zhen Dong, Bisheng Yang†
International Conference on Learning Representations (ICLR) 2025
We present a two-stage (coarse-to-fine) 3D visual grounding system by tuning Large Vision Language Model (LVLM) to accurately find targets in city-scale point clouds from text descriptions.
FreeReg':' Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators
Haiping Wang*, Yuan Liu*, Bing Wang, Yujing Sun, Zhen Dong†, Wenping Wang, Bisheng Yang†
International Conference on Learning Representations (ICLR) 2024
FreeReg extracts cross-modality features from pretrained diffusion models and monocular depth estimators for accurate zero-shot image-to-point cloud registration.
FreeReg':' Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators
Haiping Wang*, Yuan Liu*, Bing Wang, Yujing Sun, Zhen Dong†, Wenping Wang, Bisheng Yang†
International Conference on Learning Representations (ICLR) 2024
FreeReg extracts cross-modality features from pretrained diffusion models and monocular depth estimators for accurate zero-shot image-to-point cloud registration.
A novel method for registration of MLS and stereo reconstructed point clouds
Xiaochen Yang*, Haiping Wang*, Zhen Dong†, Yuan Liu, Yuhan Li, Bisheng Yang†
IEEE Transactions on Geoscience and Remote Sensing (T-GRS, IF:8.2) 2024
A fast and robust SO(2)-equivariant point cloud descriptor designed for aligning point clouds confirming 4DoF rigid-transformation such as MLS and TLS data.
A novel method for registration of MLS and stereo reconstructed point clouds
Xiaochen Yang*, Haiping Wang*, Zhen Dong†, Yuan Liu, Yuhan Li, Bisheng Yang†
IEEE Transactions on Geoscience and Remote Sensing (T-GRS, IF:8.2) 2024
A fast and robust SO(2)-equivariant point cloud descriptor designed for aligning point clouds confirming 4DoF rigid-transformation such as MLS and TLS data.
Robust Multiview Point Cloud Registration with Reliable Pose Graph Initialization and History Reweighting
Haiping Wang*, Yuan Liu*, Zhen Dong†, Yulan Guo, Yu-Shen Liu, Wenping Wang, Bisheng Yang†
The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2023
A simple and effective multiview point cloud registration method containing a sparse pose graph construction and a robust IRLS method, achieving SoTA registration performances on the 3D(Lo)Match, ScanNet, and ETH datasets (2023).
Robust Multiview Point Cloud Registration with Reliable Pose Graph Initialization and History Reweighting
Haiping Wang*, Yuan Liu*, Zhen Dong†, Yulan Guo, Yu-Shen Liu, Wenping Wang, Bisheng Yang†
The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2023
A simple and effective multiview point cloud registration method containing a sparse pose graph construction and a robust IRLS method, achieving SoTA registration performances on the 3D(Lo)Match, ScanNet, and ETH datasets (2023).
RoReg':' Pairwise Point Cloud Registration with Oriented Descriptors and Local Rotations
Haiping Wang*, Yuan Liu*, Qingyong Hu, Bing Wang, Jianguo Chen, Zhen Dong†, Yulan Guo, Wenping Wang, Bisheng Yang†
[🏆 ESI Highly Cited Paper] IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI, IF:24.3) 2023
Group-based rotation-equivariance can benefit each components of point cloud registration, including feature extraction, feature detection, feature matching, and transformation estimation. RoReg achieves SoTA registration performances on the 3D(Lo)Match and ETH datasets (2023).
RoReg':' Pairwise Point Cloud Registration with Oriented Descriptors and Local Rotations
Haiping Wang*, Yuan Liu*, Qingyong Hu, Bing Wang, Jianguo Chen, Zhen Dong†, Yulan Guo, Wenping Wang, Bisheng Yang†
[🏆 ESI Highly Cited Paper] IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI, IF:24.3) 2023
Group-based rotation-equivariance can benefit each components of point cloud registration, including feature extraction, feature detection, feature matching, and transformation estimation. RoReg achieves SoTA registration performances on the 3D(Lo)Match and ETH datasets (2023).
You Only Hypothesize Once':' Point Cloud Registration with Rotation-equivariant Descriptors
Haiping Wang*, Yuan Liu*, Zhen Dong†, Wenping Wang
ACM Multimedia (MM) 2022
Endow local descriptors of point clouds with rotation equivariance based on the icosahedral group learning, achieving SoTA registration performances on the 3D(Lo)Match, ETH, and WHU-TLS datasets (2022).
You Only Hypothesize Once':' Point Cloud Registration with Rotation-equivariant Descriptors
Haiping Wang*, Yuan Liu*, Zhen Dong†, Wenping Wang
ACM Multimedia (MM) 2022
Endow local descriptors of point clouds with rotation equivariance based on the icosahedral group learning, achieving SoTA registration performances on the 3D(Lo)Match, ETH, and WHU-TLS datasets (2022).