Hi, I’m Zongqi He. I received my Bachelor’s degree from the Department of Electrical and Electronic Engineering at The Hong Kong Polytechnic University. I was very fortunate to be advised by Prof. Kenneth K. M. Lam during my undergraduate studies. I am now a Ph.D. student at The University of Hong Kong, advised by Yifan (Evan) Peng.

My research interests include computer vision, deep learning, low-level vision, and 3D/4D reconstruction.

🔥 News

2026.07: 🎉🎉 Our paper SfM-free 3D Gaussian Splatting from Extremely Sparse View is now available online in Computers & Graphics.
2026.06: 🎉🎉 Our paper PhyGaP: Physically-Grounded Gaussians with Polarization Cues was selected as a CVPR 2026 Oral presentation.
2026.06: 🎉🎉 Our paper PhysInOne: Visual Physics Learning and Reasoning in One Suite was presented at CVPR 2026.
2026.05: 🎉🎉 Our paper Augmented Reality Integration Improves Ergonomics in Dynamic Navigation for Dental Implant Surgery received the Distinguished Student Paper Award at SID 2026.
2025.12: 🎉🎉 I joined The University of Hong Kong (HKU) as a Ph.D. student.

🚧 Ongoing Projects

RestoreAvatars: Developing a high-fidelity head-avatar reconstruction method that is robust to degraded facial sequences from old films.
ConsistNav: Improving action consistency in zero-shot object navigation through semantic executive control.
Continuous-Time Gaussian Motion Adapters: Adapting frozen dynamic-scene reconstruction models with continuous-time Gaussian motion representations.
Illumination-Aware Colonoscopic Visual Odometry: Exploring self-supervised visual odometry for colonoscopy using illumination-aware 3D Gaussian splatting.
Generative Sparse-View Synthesis: Investigating collaborative learning for sparse-view synthesis based on generative 3D Gaussian splatting.

CVPR 2026 Oral

Jiale Wu, Xiaoyang Bai, Zongqi He, Weiwei Xu, Yifan (Evan) Peng.

CVPR 2026

Siyuan Zhou*, Hejun Wang*, Hu Cheng*, Jinxi Li*, DataTeam (including Zongqi He), et al.

🏆 Distinguished Student Paper Award

Pui Hang Leung, Feng Wang, Zhenyang Li, Zongqi He, Yifan Peng, Wei-fa Yang.

Journal of the Society for Information Display, 34(5): 428-435, 2026.

Computers & Graphics 2026

Zongqi He, Hanmin Li, Kin-Chung Chan, Yushen Zuo, Hao Xie, Zhe Xiao, Jun Xiao, Xiaoyang Bai, Yifan Peng, Kin-Man Lam.

ICASSP 2025

Zongqi He, Zhe Xiao, Kin-Chung Chan, Yushen Zuo, Jun Xiao, Kin-Man Lam.

AI4VA@ECCV 2024

Towards Multi-View Consistent Style Transfer with One-Step Diffusion via Vision Conditioning

Yushen Zuo, Jun Xiao, Kin-Chung Chan, Rongkang Dong, Cuixin Yang, Zongqi He, Hao Xie, Kin-Man Lam

2025.03 NTIRE 2025 Challenge on Night Photography Rendering - 5th place.
2025.03 NTIRE 2025 Challenge on Ambient Light Normalization - 6th place.
2025.03 NTIRE 2025 Challenge on Restore Any Image Model (RAIM) in the Wild - Track 1 - 3rd place.
2024.08 AIM 2024 Challenge on Sparse Neural Rendering - Track 1 - 3 views - 3rd place.
2024.08 AIM 2024 Challenge on Sparse Neural Rendering - Track 2 - 9 views - 3rd place.
2024.08 AIM 2024 Challenge on Efficient Video Super-Resolution for AV1 Compressed Content - 2nd place.
2022-2024 Dean’s Honours List (two years)
2023 HKSAR Government Talent Development Scholarship

Computer Vision, 3D Reconstruction, and Image Processing — The Hong Kong Polytechnic University, advised by Jun Xiao and Kenneth Lam
- Developed SIDGaussian for sparse-view novel view synthesis with semantic and local-depth regularization; this work was accepted by ICASSP 2025.
- Built a one-step, multi-view-consistent diffusion framework with vision conditioning and LoRA-based adaptation for efficient style transfer.
- Improved retina OCT image denoising and classification with a frequency-domain loss and an enhanced classification backbone.
PhysInOne: Physics-Based Dataset — Research Assistant, The Hong Kong Polytechnic University
- Contributed to the development of PhysInOne, a large-scale suite for visual physics learning and reasoning.
- Built foundational Unreal Engine and Unity skills for scene construction and animation workflows, and created UE materials and procedural workflows for scalable scene creation.