Jiakai Sun

I am currently pursuing a master's degree in Software Engineering at the College of Computer Science and Technology (CCST) of Zhejiang University (ZJU). I earned my bachelor's degree from CCST, ZJU, in 2022.

My research interests are centered around Computer Graphics and Computer Vision. I am particularly focused on Neural Rendering, Inverse Rendering, and 3D Vision, exploring the cutting-edge developments within these domains.

I will complete my Master's in March 2025 and am actively seeking job opportunities. Please contact me if you're interested in my background 😊

Email  /  GitHub  /  Google Scholar  

Selected Publications

3DGStream: On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos

We introduce a system designed for efficient Free-Viewpoint Videos streaming of real-world dynamic scenes, achieving fast on-the-fly per-frame reconstruction within 12 seconds and real-time rendering at 200 FPS.

Joint Optimization of Triangle Mesh, Material, and Light from Neural Fields with Neural Radiance Cache

Our framework merges neural fields and radiance caching to account for global illuination in single-bounce differentiable Monte Carlo ray tracing, achieving high-quality mesh and material reconstruction.

VGOS: Voxel Grid Optimization for View Synthesis from Sparse Inputs

We present a novel approach that rapidly reconstructs high-quality radiance fields from as few as 3-10 views in just 3-5 minutes, overcoming the limitations of voxel grid overfitting and artifacts in sparse input scenarios through an incremental training strategy and regularization techniques.

Other Publications

Single-Mask Inpainting for Voxel-based Neural Radiance Fields

Rethinking Video Deblurring with Wavelet-Aware Dynamic Transformer and Diffusion Model

PNeSM: Arbitrary 3D Scene Stylization via Prompt-Based Neural Style Mapping

ArtBank: Artistic Style Transfer with Pre-trained Diffusion Model and Implicit Style Prompt Bank

Attack Deterministic Conditional Image Generative Models for Diverse and Controllable Generation

DuDoINet: Dual-Domain Implicit Network for Multi-Modality MR Image Arbitrary-scale Super-Resolution

Self-Reference Image Super-Resolution via Pre-trained Diffusion Large Model and Window Adjustable Transformer

Rethinking fast fourier convolution in image inpainting

Rethinking Multi-Contrast MRI Super-Resolution: Rectangle-Window Cross-Attention Transformer and Arbitrary-Scale Upsampling

Source code from Leonid Keselman's website, which is designed from Jon Barron's website.