Jiarui Meng （孟佳锐）

I am currently an AI Engineer at Huawei. I received my Master's degree in Computer Applied Technology from Peking University, where I was advised by Prof. Jian Zhang and Prof. Siwei Ma. Prior to that, I obtained my Bachelor's degree from Harbin Institute of Technology.

I am interested in 3D Reconstruction, 3D Vision and 3D-LLMs.

Email  /  Scholar  

Publications

We propose SLARM, a feed-forward model that unifies dynamic scene reconstruction, semantic understanding, and real-time streaming inference. SLARM captures complex, non-uniform motion through higher-order motion modeling, trained solely on differentiable renderings without any flow supervision.

We propose SecureGS, a secure and efficient 3DGS steganography framework inspired by Scaffold-GS's anchor point design and neural decoding. SecureGS uses a hybrid decoupled Gaussian encryption mechanism to embed offsets, scales, rotations, and RGB attributes of the hidden 3D Gaussian points in anchor point features, retrievable only by authorized users through privacy-preserving neural networks.

We propose InstanceGaussian, a novel 3D scene understanding approach that jointly learns appearance and semantic features through a semantic-scaffold Gaussian representation, progressive joint training, and bottom-up instance aggregation, achieving state-of-the-art performance in category-agnostic open-vocabulary 3D point-level segmentation.

To address the security challenges in 3D Gaussian Splatting, we propose GS-Hider, a novel steganography framework that invisibly embeds multimodal messages into point clouds via coupled secured features, achieving high fidelity, robustness, and security without compromising rendering quality.

We introduce Mirror-3DGS, a novel framework that integrates mirror attributes and plane imaging principles into 3D Gaussian Splatting to accurately model physical reflections, achieving superior real-time rendering fidelity in mirror regions compared to state-of-the-art methods.

This paper introduces OpenGaussian, a method for 3D Gaussian Splatting that achieves robust 3D point-level open vocabulary understanding by training instance features with 3D consistency, utilizing a two-stage codebook for discretization, and establishing instance-level 3D-2D feature associations.

We propose HiCoM, an efficient framework for dynamic scene reconstruction that leverages a compact 3D Gaussian Splatting representation and a Hierarchical Coherent Motion mechanism to achieve superior training speed, reduced storage, and robust real-time rendering performance.

Internships

Selected Honors

Service