I am a PhD student at Shanghai Jiao Tong University, working at the intersection of 3D Vision and Neural Data Compression. My research focuses on making 3D representations (like Gaussian Splatting) more efficient and robust, and leveraging state-of-the-art architectures (like Mamba) for image and video compression.

• [2026.01] 🎉 SurfSplat accepted to ICLR 2026! Catch me in Rio de Janeiro this April.
• [2025.10] 🚀 H3D-DGS accepted to NeurIPS 2025.

SurfSplat: Conquering Feedforward 2D Gaussian Splatting ICLR 2026 | Project Page

• Introducing surface continuity priors to 2DGS. No more artifacts—finally, high-fidelity 3DGS that holds up even under extreme zoom.

H3D-DGS: Heterogeneous 3D Motion Representation NeurIPS 2025 | Project Page

• Tackling intense motion in deformable scenes. We combine neural networks with graphics priors to accelerate training and stabilize deformation.

Content-Aware Mamba (CMIC) ICLR 2026 * Moving beyond rigid raster scans. We introduce a content-adaptive token permutation strategy for Mamba-style SSMs, achieving over 21% BD-rate savings vs VTM-21.0.

Knowledge Distillation for Learned Image Compression ICCV 2025(oral) * Knowledge distillation meets learned image compression! We’ve developed a streamlined framework that delivers a smaller, faster model with significantly reduced computational overhead, all while maintaining state-of-the-art performance.

Implicit-Explicit Hybrid Multi-View Codec TIP 2025 * Solving the VR data bottleneck. A framework that fuses traditional 2D codecs with Implicit Neural Representations (INR) to outperform the MIV standard.

• Languages: C++ (TA for Undergraduate C++), Python, Cython.
• Frameworks: PyTorch.
• Expertise: 3D Reconstruction, 3DGS, Rigging.